fix(mcp): resolve server connection errors and improve error handling

pull/819/head
DP37 3 months ago committed by ascender1729
parent f61ada7928
commit 616c5757b0

@ -23,11 +23,11 @@ args = "python -m unittest tests/tools/test_mcp_integration.py -v"
[[workflows.workflow]]
name = "Run MCP Demo"
author = 13983571
mode = "sequential"
mode = "parallel"
[[workflows.workflow.tasks]]
task = "shell.exec"
args = "python examples/mcp_example/mock_math_server.py & "
args = "python examples/mcp_example/mock_math_server.py"
[[workflows.workflow.tasks]]
task = "shell.exec"

@ -0,0 +1,100 @@
The root of that “unhandled errors in a TaskGroup (1 subexception)” is simply that your clients `MCPServerSse.connect()` is failing under the hood (most likely a connection/refused or pathnotfound error) and AnyIO is wrapping it in a TaskGroup exception. You dont see the real cause because it gets hidden by AnyIOs TaskGroup. Heres how to unmask it and fix it:
---
## 1. Diagnose the real error
Wrap the connect call and print the underlying exception:
```python
async def _test_connect():
server = MCPServerSse(get_server_params())
try:
await server.connect()
await server.cleanup()
return True
except Exception as e:
# Print the actual cause
import traceback; traceback.print_exc()
return False
print(asyncio.run(_test_connect()))
```
Youll probably see a **connection refused** or **404 on /sse** in the stack trace.
---
## 2. Ensure client and server agree on your SSE endpoint
By default FastMCP serves its SSE stream at `/sse` and messages on `/messages`. If you only pass `url="http://127.0.0.1:8000"` the client will try whatever its default path is (often `/events` or `/stream`). You need to be explicit:
```python
from swarms.tools.mcp_integration import MCPServerSseParams
def get_server_params():
return MCPServerSseParams(
url="http://127.0.0.1:8000",
sse_path="/sse", # <— tell it exactly where the SSE lives
messages_path="/messages", # <— if your server uses /messages for POSTs
headers={
"Content-Type": "application/json",
"Accept": "text/event-stream",
},
timeout=15.0,
sse_read_timeout=60.0,
require_session_id=False, # match your servers require_session_id
)
```
---
## 3. Dont manually call `MCPServerSse` unless you need to
Your `test_server_connection()` can more reliably just do a raw HTTP(S) healthcheck:
```python
def test_server_connection():
health_url = get_server_params().url + get_server_params().sse_path
try:
r = httpx.get(health_url,
headers={"Accept":"text/event-stream"},
timeout=5.0)
if r.status_code == 200:
logger.info("✅ SSE endpoint is up")
return True
else:
logger.error(f"❌ Unexpected status {r.status_code}")
return False
except Exception as e:
logger.error(f"❌ Connection to SSE endpoint failed: {e}")
return False
```
That way you see immediately if the server is refusing connections or returning 404.
---
## 4. Align your Agent configuration
Once youve verified the raw GET to `http://127.0.0.1:8000/sse` is 200, your Agent should work with exactly the same params:
```python
math_agent = Agent(
agent_name="Math Assistant",
agent_description="Friendly math calculator",
system_prompt=MATH_AGENT_PROMPT,
max_loops=1,
model_name="gpt-3.5-turbo",
verbose=True,
mcp_servers=[ get_server_params() ]
)
```
Now when you do `math_agent.run("add 3 and 4")`, the SSE handshake will succeed and youll no longer see that TaskGroup error.
---
### TL;DR
1. **Print the real exception** behind the TaskGroup to see “connection refused” or “404.”
2. **Explicitly set** `sse_path="/sse"` (and `messages_path`) in `MCPServerSseParams`.
3. **Healthcheck** with a simple `httpx.get("…/sse")` instead of `server.connect()`.
4. Pass those same params straight into your `Agent`.
Once your client is pointing at the exact SSE URL your FastMCP server is serving, the Agent will connect cleanly and youll be back to doing math instead of wrestling TaskGroup errors.

@ -0,0 +1,397 @@
from swarms import Agent
from swarms.tools.mcp_integration import MCPServerSseParams, MCPServerSse, mcp_flow_get_tool_schema
from loguru import logger
import sys
import asyncio
import json
import httpx
import time
# Configure logging for more detailed output
logger.remove()
logger.add(sys.stdout,
level="DEBUG",
format="{time} | {level} | {module}:{function}:{line} - {message}")
# Relaxed prompt that doesn't enforce strict JSON formatting
# Create server parameters
def get_server_params():
"""Get the MCP server connection parameters."""
return MCPServerSseParams(
url=
"http://127.0.0.1:8000", # Use 127.0.0.1 instead of localhost/0.0.0.0
headers={
"Content-Type": "application/json",
"Accept": "text/event-stream"
},
timeout=15.0, # Longer timeout
sse_read_timeout=60.0 # Longer read timeout
)
def initialize_math_system():
"""Initialize the math agent with MCP server configuration."""
# Create the agent with the MCP server configuration
math_agent = Agent(agent_name="Math Assistant",
agent_description="Friendly math calculator",
system_prompt=MATH_AGENT_PROMPT,
max_loops=1,
mcp_servers=[get_server_params()],
model_name="gpt-3.5-turbo",
verbose=True)
return math_agent
# Function to get list of available tools from the server
async def get_tools_list():
"""Fetch and format the list of available tools from the server."""
try:
server_params = get_server_params()
tools = await mcp_flow_get_tool_schema(server_params)
if not tools:
return "No tools are currently available on the server."
# Format the tools information
tools_info = "Available tools:\n"
for tool in tools:
tools_info += f"\n- {tool.name}: {tool.description or 'No description'}\n"
if tool.parameters and hasattr(tool.parameters, 'properties'):
tools_info += " Parameters:\n"
for param_name, param_info in tool.parameters.properties.items(
):
param_type = param_info.get('type', 'unknown')
param_desc = param_info.get('description',
'No description')
tools_info += f" - {param_name} ({param_type}): {param_desc}\n"
return tools_info
except Exception as e:
logger.error(f"Failed to get tools list: {e}")
return f"Error retrieving tools list: {str(e)}"
# Function to test server connection
def test_server_connection():
"""Test if the server is reachable and responsive."""
try:
# Create a short-lived connection to check server
server = MCPServerSse(get_server_params())
# Try connecting (this is synchronous)
asyncio.run(server.connect())
asyncio.run(server.cleanup())
logger.info("✅ Server connection test successful")
return True
except Exception as e:
logger.error(f"❌ Server connection test failed: {e}")
return False
# Manual math operation handler as ultimate fallback
def manual_math(query):
"""Parse and solve a math problem without using the server."""
query = query.lower()
# Check if user is asking for available tools/functions
if "list" in query and ("tools" in query or "functions" in query
or "operations" in query):
return """
Available tools:
1. add - Add two numbers together (e.g., "add 3 and 4")
2. multiply - Multiply two numbers together (e.g., "multiply 5 and 6")
3. divide - Divide the first number by the second (e.g., "divide 10 by 2")
"""
try:
if "add" in query or "plus" in query or "sum" in query:
# Extract numbers using a simple approach
numbers = [int(s) for s in query.split() if s.isdigit()]
if len(numbers) >= 2:
result = numbers[0] + numbers[1]
return f"The sum of {numbers[0]} and {numbers[1]} is {result}"
elif "multiply" in query or "times" in query or "product" in query:
numbers = [int(s) for s in query.split() if s.isdigit()]
if len(numbers) >= 2:
result = numbers[0] * numbers[1]
return f"The product of {numbers[0]} and {numbers[1]} is {result}"
elif "divide" in query or "quotient" in query:
numbers = [int(s) for s in query.split() if s.isdigit()]
if len(numbers) >= 2:
if numbers[1] == 0:
return "Cannot divide by zero"
result = numbers[0] / numbers[1]
return f"{numbers[0]} divided by {numbers[1]} is {result}"
return "I couldn't parse your math request. Try something like 'add 3 and 4'."
except Exception as e:
logger.error(f"Manual math error: {e}")
return f"Error performing calculation: {str(e)}"
def main():
try:
logger.info("Initializing math system...")
# Test server connection first
server_available = test_server_connection()
if server_available:
math_agent = initialize_math_system()
print("\nMath Calculator Ready! (Server connection successful)")
else:
print(
"\nServer connection failed - using fallback calculator mode")
math_agent = None
print("Ask me any math question!")
print("Examples: 'what is 5 plus 3?' or 'can you multiply 4 and 6?'")
print("Type 'list tools' to see available operations")
print("Type 'exit' to quit\n")
while True:
try:
query = input("What would you like to calculate? ").strip()
if not query:
continue
if query.lower() == 'exit':
break
# Handle special commands
if query.lower() in ('list tools', 'show tools',
'available tools', 'what tools'):
if server_available:
# Get tools list from server
tools_info = asyncio.run(get_tools_list())
print(f"\n{tools_info}\n")
else:
# Use manual fallback
print(manual_math("list tools"))
continue
logger.info(f"Processing query: {query}")
# First try the agent if available
if math_agent and server_available:
try:
result = math_agent.run(query)
print(f"\nResult: {result}\n")
continue
except Exception as e:
logger.error(f"Agent error: {e}")
print("Agent encountered an error, trying fallback...")
# If agent fails or isn't available, use manual calculator
result = manual_math(query)
print(f"\nCalculation result: {result}\n")
except KeyboardInterrupt:
print("\nGoodbye!")
break
except Exception as e:
logger.error(f"Error processing query: {e}")
print(f"Sorry, there was an error: {str(e)}")
except Exception as e:
logger.error(f"System initialization error: {e}")
print(f"Failed to start the math system: {str(e)}")
if __name__ == "__main__":
main() "from fastmcp import FastMCP
from loguru import logger
import time
import json
# Create the MCP server with detailed debugging
mcp = FastMCP(
host="0.0.0.0", # Bind to all interfaces
port=8000,
transport="sse",
require_session_id=False,
cors_allowed_origins=["*"], # Allow connections from any origin
debug=True # Enable debug mode for more verbose output
)
# Add a more flexible parsing approach
def parse_input(input_str):
"""Parse input that could be JSON or natural language."""
try:
# First try to parse as JSON
return json.loads(input_str)
except json.JSONDecodeError:
# If not JSON, try to parse natural language
input_lower = input_str.lower()
# Parse for addition
if "add" in input_lower or "plus" in input_lower or "sum" in input_lower:
# Extract numbers - very simple approach
numbers = [int(s) for s in input_lower.split() if s.isdigit()]
if len(numbers) >= 2:
return {"a": numbers[0], "b": numbers[1]}
# Parse for multiplication
if "multiply" in input_lower or "times" in input_lower or "product" in input_lower:
numbers = [int(s) for s in input_lower.split() if s.isdigit()]
if len(numbers) >= 2:
return {"a": numbers[0], "b": numbers[1]}
# Parse for division
if "divide" in input_lower or "quotient" in input_lower:
numbers = [int(s) for s in input_lower.split() if s.isdigit()]
if len(numbers) >= 2:
return {"a": numbers[0], "b": numbers[1]}
# Could not parse successfully
return None
# Define tools with more flexible input handling
@mcp.tool()
def add(input_str=None, a=None, b=None):
"""Add two numbers. Can accept JSON parameters or natural language.
Args:
input_str (str, optional): Natural language input to parse
a (int, optional): First number if provided directly
b (int, optional): Second number if provided directly
Returns:
str: A message containing the sum
"""
logger.info(f"Add tool called with input_str={input_str}, a={a}, b={b}")
# If we got a natural language string instead of parameters
if input_str and not (a is not None and b is not None):
parsed = parse_input(input_str)
if parsed:
a = parsed.get("a")
b = parsed.get("b")
# Validate we have what we need
if a is None or b is None:
return "Sorry, I couldn't understand the numbers to add"
try:
a = int(a)
b = int(b)
result = a + b
return f"The sum of {a} and {b} is {result}"
except ValueError:
return "Please provide valid numbers for addition"
@mcp.tool()
def multiply(input_str=None, a=None, b=None):
"""Multiply two numbers. Can accept JSON parameters or natural language.
Args:
input_str (str, optional): Natural language input to parse
a (int, optional): First number if provided directly
b (int, optional): Second number if provided directly
Returns:
str: A message containing the product
"""
logger.info(
f"Multiply tool called with input_str={input_str}, a={a}, b={b}")
# If we got a natural language string instead of parameters
if input_str and not (a is not None and b is not None):
parsed = parse_input(input_str)
if parsed:
a = parsed.get("a")
b = parsed.get("b")
# Validate we have what we need
if a is None or b is None:
return "Sorry, I couldn't understand the numbers to multiply"
try:
a = int(a)
b = int(b)
result = a * b
return f"The product of {a} and {b} is {result}"
except ValueError:
return "Please provide valid numbers for multiplication"
@mcp.tool()
def divide(input_str=None, a=None, b=None):
"""Divide two numbers. Can accept JSON parameters or natural language.
Args:
input_str (str, optional): Natural language input to parse
a (int, optional): Numerator if provided directly
b (int, optional): Denominator if provided directly
Returns:
str: A message containing the division result or an error message
"""
logger.info(f"Divide tool called with input_str={input_str}, a={a}, b={b}")
# If we got a natural language string instead of parameters
if input_str and not (a is not None and b is not None):
parsed = parse_input(input_str)
if parsed:
a = parsed.get("a")
b = parsed.get("b")
# Validate we have what we need
if a is None or b is None:
return "Sorry, I couldn't understand the numbers to divide"
try:
a = int(a)
b = int(b)
if b == 0:
logger.warning("Division by zero attempted")
return "Cannot divide by zero"
result = a / b
return f"{a} divided by {b} is {result}"
except ValueError:
return "Please provide valid numbers for division"
if __name__ == "__main__":
try:
logger.info("Starting math server on http://0.0.0.0:8000")
print("Math MCP Server is running. Press Ctrl+C to stop.")
print(
"Server is configured to accept both JSON and natural language input"
)
# Add a small delay to ensure logging is complete before the server starts
time.sleep(0.5)
# Run the MCP server
mcp.run()
except KeyboardInterrupt:
logger.info("Server shutdown requested")
print("\nShutting down server...")
except Exception as e:
logger.error(f"Server error: {e}")
raise
" server is runnig poeroperly "2025-04-20 17:35:01.251 | INFO | __main__:<module>:161 - Starting math server on http://0.0.0.0:8000
Math MCP Server is running. Press Ctrl+C to stop.
Server is configured to accept both JSON and natural language input
[04/20/25 17:35:01] INFO Starting server "FastMCP"... " butwhy im getting these errore "2025-04-20T17:35:04.174629+0000 | INFO | mcp_client:main:159 - Initializing math system...
2025-04-20T17:35:04.203591+0000 | ERROR | mcp_integration:connect:89 - Error initializing MCP server: unhandled errors in a TaskGroup (1 sub-exception)
2025-04-20T17:35:04.204437+0000 | ERROR | mcp_client:test_server_connection:110 - ❌ Server connection test failed: unhandled errors in a TaskGroup (1 sub-exception)
Server connection failed - using fallback calculator mode
Ask me any math question!
Examples: 'what is 5 plus 3?' or 'can you multiply 4 and 6?'
Type 'list tools' to see available operations
Type 'exit' to quit
What would you like to calculate? "

@ -1,60 +1,158 @@
from swarms import Agent
from swarms.tools.mcp_integration import MCPServerSseParams
from swarms.tools.mcp_integration import MCPServerSseParams, MCPServerSse, mcp_flow_get_tool_schema
from loguru import logger
import sys
import asyncio
import json
import httpx
import time
# Configure logging for more detailed output
logger.remove()
logger.add(sys.stdout,
level="DEBUG",
format="{time} | {level} | {module}:{function}:{line} - {message}")
# Relaxed prompt that doesn't enforce strict JSON formatting
# Create server parameters
def get_server_params():
"""Get the MCP server connection parameters."""
return MCPServerSseParams(
url=
"http://127.0.0.1:8000", # Use 127.0.0.1 instead of localhost/0.0.0.0
headers={
"Content-Type": "application/json",
"Accept": "text/event-stream"
},
timeout=15.0, # Longer timeout
sse_read_timeout=60.0 # Longer read timeout
)
# Comprehensive math prompt that encourages proper JSON formatting
MATH_AGENT_PROMPT = """
You are a helpful math calculator assistant.
Your role is to understand natural language math requests and perform calculations.
When asked to perform calculations:
def initialize_math_system():
"""Initialize the math agent with MCP server configuration."""
# Create the agent with the MCP server configuration
math_agent = Agent(agent_name="Math Assistant",
agent_description="Friendly math calculator",
system_prompt=MATH_AGENT_PROMPT,
max_loops=1,
mcp_servers=[get_server_params()],
model_name="gpt-3.5-turbo",
verbose=True)
1. Determine the operation (add, multiply, or divide)
2. Extract the numbers from the request
3. Use the appropriate math operation tool
return math_agent
FORMAT YOUR TOOL CALLS AS JSON with this format:
{"tool_name": "add", "a": <first_number>, "b": <second_number>}
or
{"tool_name": "multiply", "a": <first_number>, "b": <second_number>}
or
{"tool_name": "divide", "a": <first_number>, "b": <second_number>}
Always respond with a tool call in JSON format first, followed by a brief explanation.
"""
# Function to get list of available tools from the server
async def get_tools_list():
"""Fetch and format the list of available tools from the server."""
try:
server_params = get_server_params()
tools = await mcp_flow_get_tool_schema(server_params)
if not tools:
return "No tools are currently available on the server."
# Format the tools information
tools_info = "Available tools:\n"
for tool in tools:
tools_info += f"\n- {tool.name}: {tool.description or 'No description'}\n"
if tool.parameters and hasattr(tool.parameters, 'properties'):
tools_info += " Parameters:\n"
for param_name, param_info in tool.parameters.properties.items(
):
param_type = param_info.get('type', 'unknown')
param_desc = param_info.get('description',
'No description')
tools_info += f" - {param_name} ({param_type}): {param_desc}\n"
return tools_info
except Exception as e:
logger.error(f"Failed to get tools list: {e}")
return f"Error retrieving tools list: {str(e)}"
def initialize_math_system():
"""Initialize the math agent with MCP server configuration."""
# Configure the MCP server connection
math_server = MCPServerSseParams(
url="http://0.0.0.0:8000",
headers={"Content-Type": "application/json"},
timeout=5.0,
sse_read_timeout=30.0
)
# Create the agent with the MCP server configuration
math_agent = Agent(
agent_name="Math Assistant",
agent_description="Friendly math calculator",
system_prompt=MATH_AGENT_PROMPT,
max_loops=1,
mcp_servers=[math_server], # Pass MCP server config as a list
model_name="gpt-3.5-turbo",
verbose=True # Enable verbose mode to see more details
)
# Function to test server connection
def test_server_connection():
"""Test if the server is reachable and responsive."""
try:
# Create a short-lived connection to check server
server = MCPServerSse(get_server_params())
# Try connecting (this is synchronous)
asyncio.run(server.connect())
asyncio.run(server.cleanup())
logger.info("✅ Server connection test successful")
return True
except Exception as e:
logger.error(f"❌ Server connection test failed: {e}")
return False
# Manual math operation handler as ultimate fallback
def manual_math(query):
"""Parse and solve a math problem without using the server."""
query = query.lower()
# Check if user is asking for available tools/functions
if "list" in query and ("tools" in query or "functions" in query
or "operations" in query):
return """
Available tools:
1. add - Add two numbers together (e.g., "add 3 and 4")
2. multiply - Multiply two numbers together (e.g., "multiply 5 and 6")
3. divide - Divide the first number by the second (e.g., "divide 10 by 2")
"""
try:
if "add" in query or "plus" in query or "sum" in query:
# Extract numbers using a simple approach
numbers = [int(s) for s in query.split() if s.isdigit()]
if len(numbers) >= 2:
result = numbers[0] + numbers[1]
return f"The sum of {numbers[0]} and {numbers[1]} is {result}"
elif "multiply" in query or "times" in query or "product" in query:
numbers = [int(s) for s in query.split() if s.isdigit()]
if len(numbers) >= 2:
result = numbers[0] * numbers[1]
return f"The product of {numbers[0]} and {numbers[1]} is {result}"
elif "divide" in query or "quotient" in query:
numbers = [int(s) for s in query.split() if s.isdigit()]
if len(numbers) >= 2:
if numbers[1] == 0:
return "Cannot divide by zero"
result = numbers[0] / numbers[1]
return f"{numbers[0]} divided by {numbers[1]} is {result}"
return "I couldn't parse your math request. Try something like 'add 3 and 4'."
except Exception as e:
logger.error(f"Manual math error: {e}")
return f"Error performing calculation: {str(e)}"
return math_agent
def main():
try:
logger.info("Initializing math system...")
math_agent = initialize_math_system()
print("\nMath Calculator Ready!")
# Test server connection first
server_available = test_server_connection()
if server_available:
math_agent = initialize_math_system()
print("\nMath Calculator Ready! (Server connection successful)")
else:
print(
"\nServer connection failed - using fallback calculator mode")
math_agent = None
print("Ask me any math question!")
print("Examples: 'what is 5 plus 3?' or 'can you multiply 4 and 6?'")
print("Type 'list tools' to see available operations")
print("Type 'exit' to quit\n")
while True:
@ -65,9 +163,33 @@ def main():
if query.lower() == 'exit':
break
# Handle special commands
if query.lower() in ('list tools', 'show tools',
'available tools', 'what tools'):
if server_available:
# Get tools list from server
tools_info = asyncio.run(get_tools_list())
print(f"\n{tools_info}\n")
else:
# Use manual fallback
print(manual_math("list tools"))
continue
logger.info(f"Processing query: {query}")
result = math_agent.run(query)
print(f"\nResult: {result}\n")
# First try the agent if available
if math_agent and server_available:
try:
result = math_agent.run(query)
print(f"\nResult: {result}\n")
continue
except Exception as e:
logger.error(f"Agent error: {e}")
print("Agent encountered an error, trying fallback...")
# If agent fails or isn't available, use manual calculator
result = manual_math(query)
print(f"\nCalculation result: {result}\n")
except KeyboardInterrupt:
print("\nGoodbye!")
@ -80,5 +202,6 @@ def main():
logger.error(f"System initialization error: {e}")
print(f"Failed to start the math system: {str(e)}")
if __name__ == "__main__":
main()

@ -1,70 +1,168 @@
from fastmcp import FastMCP
from loguru import logger
import time
# Create the MCP server
mcp = FastMCP(host="0.0.0.0",
port=8000,
transport="sse",
require_session_id=False)
# Define tools with proper type hints and docstrings
import json
# Create the MCP server with detailed debugging
mcp = FastMCP(
host="0.0.0.0", # Bind to all interfaces
port=8000,
transport="sse",
require_session_id=False,
cors_allowed_origins=["*"], # Allow connections from any origin
debug=True # Enable debug mode for more verbose output
)
# Add a more flexible parsing approach
def parse_input(input_str):
"""Parse input that could be JSON or natural language."""
try:
# First try to parse as JSON
return json.loads(input_str)
except json.JSONDecodeError:
# If not JSON, try to parse natural language
input_lower = input_str.lower()
# Parse for addition
if "add" in input_lower or "plus" in input_lower or "sum" in input_lower:
# Extract numbers - very simple approach
numbers = [int(s) for s in input_lower.split() if s.isdigit()]
if len(numbers) >= 2:
return {"a": numbers[0], "b": numbers[1]}
# Parse for multiplication
if "multiply" in input_lower or "times" in input_lower or "product" in input_lower:
numbers = [int(s) for s in input_lower.split() if s.isdigit()]
if len(numbers) >= 2:
return {"a": numbers[0], "b": numbers[1]}
# Parse for division
if "divide" in input_lower or "quotient" in input_lower:
numbers = [int(s) for s in input_lower.split() if s.isdigit()]
if len(numbers) >= 2:
return {"a": numbers[0], "b": numbers[1]}
# Could not parse successfully
return None
# Define tools with more flexible input handling
@mcp.tool()
def add(a: int, b: int) -> str:
"""Add two numbers.
def add(input_str=None, a=None, b=None):
"""Add two numbers. Can accept JSON parameters or natural language.
Args:
a (int): First number
b (int): Second number
input_str (str, optional): Natural language input to parse
a (int, optional): First number if provided directly
b (int, optional): Second number if provided directly
Returns:
str: A message containing the sum
"""
logger.info(f"Adding {a} and {b}")
result = a + b
return f"The sum of {a} and {b} is {result}"
logger.info(f"Add tool called with input_str={input_str}, a={a}, b={b}")
# If we got a natural language string instead of parameters
if input_str and not (a is not None and b is not None):
parsed = parse_input(input_str)
if parsed:
a = parsed.get("a")
b = parsed.get("b")
# Validate we have what we need
if a is None or b is None:
return "Sorry, I couldn't understand the numbers to add"
try:
a = int(a)
b = int(b)
result = a + b
return f"The sum of {a} and {b} is {result}"
except ValueError:
return "Please provide valid numbers for addition"
@mcp.tool()
def multiply(a: int, b: int) -> str:
"""Multiply two numbers.
def multiply(input_str=None, a=None, b=None):
"""Multiply two numbers. Can accept JSON parameters or natural language.
Args:
a (int): First number
b (int): Second number
input_str (str, optional): Natural language input to parse
a (int, optional): First number if provided directly
b (int, optional): Second number if provided directly
Returns:
str: A message containing the product
"""
logger.info(f"Multiplying {a} and {b}")
result = a * b
return f"The product of {a} and {b} is {result}"
logger.info(
f"Multiply tool called with input_str={input_str}, a={a}, b={b}")
# If we got a natural language string instead of parameters
if input_str and not (a is not None and b is not None):
parsed = parse_input(input_str)
if parsed:
a = parsed.get("a")
b = parsed.get("b")
# Validate we have what we need
if a is None or b is None:
return "Sorry, I couldn't understand the numbers to multiply"
try:
a = int(a)
b = int(b)
result = a * b
return f"The product of {a} and {b} is {result}"
except ValueError:
return "Please provide valid numbers for multiplication"
@mcp.tool()
def divide(a: int, b: int) -> str:
"""Divide two numbers.
def divide(input_str=None, a=None, b=None):
"""Divide two numbers. Can accept JSON parameters or natural language.
Args:
a (int): Numerator
b (int): Denominator
input_str (str, optional): Natural language input to parse
a (int, optional): Numerator if provided directly
b (int, optional): Denominator if provided directly
Returns:
str: A message containing the division result or an error message
"""
logger.info(f"Dividing {a} by {b}")
if b == 0:
logger.warning("Division by zero attempted")
return "Cannot divide by zero"
result = a / b
return f"{a} divided by {b} is {result}"
logger.info(f"Divide tool called with input_str={input_str}, a={a}, b={b}")
# If we got a natural language string instead of parameters
if input_str and not (a is not None and b is not None):
parsed = parse_input(input_str)
if parsed:
a = parsed.get("a")
b = parsed.get("b")
# Validate we have what we need
if a is None or b is None:
return "Sorry, I couldn't understand the numbers to divide"
try:
a = int(a)
b = int(b)
if b == 0:
logger.warning("Division by zero attempted")
return "Cannot divide by zero"
result = a / b
return f"{a} divided by {b} is {result}"
except ValueError:
return "Please provide valid numbers for division"
if __name__ == "__main__":
try:
logger.info("Starting math server on http://0.0.0.0:8000")
print("Math MCP Server is running. Press Ctrl+C to stop.")
print(
"Server is configured to accept both JSON and natural language input"
)
# Add a small delay to ensure logging is complete before the server starts
time.sleep(0.5)

@ -1,24 +1,25 @@
# Agent prompts for MCP testing and interactions
# Keeping the original format that already has JSON formatting
MATH_AGENT_PROMPT = """You are a helpful math calculator assistant.
MATH_AGENT_PROMPT = """
You are a helpful math calculator assistant.
Your role is to understand natural language math requests and perform calculations.
When asked to perform calculations:
1. Determine the operation (add, multiply, or divide)
2. Extract the numbers from the request
3. Use the appropriate math operation tool
Format your tool calls as JSON with the tool_name and parameters.
3. Call the appropriate operation
Example:
User: "what is 5 plus 3?"
You: Using the add operation for 5 and 3
{"tool_name": "add", "a": 5, "b": 3}
You can use these tools:
- add: Add two numbers together
- multiply: Multiply two numbers together
- divide: Divide the first number by the second number
User: "multiply 4 times 6"
You: Using multiply for 4 and 6
{"tool_name": "multiply", "a": 4, "b": 6}
"""
If the user asks for a list of available tools or functions, tell them about the above operations.
Just tell me which operation to perform and what numbers to use in natural language.
No need for strict JSON formatting - I'll handle the tool calling for you.
"""
FINANCE_AGENT_PROMPT = """You are a financial analysis agent with access to stock market data services.
Key responsibilities:
1. Interpret financial queries and determine required data

File diff suppressed because it is too large Load Diff

@ -1,320 +1,311 @@
from __future__ import annotations
import abc
import asyncio
from contextlib import AbstractAsyncContextManager, AsyncExitStack
from pathlib import Path
from typing import Any, Dict, List, Optional, Literal, Union
from typing_extensions import NotRequired, TypedDict
from anyio.streams.memory import MemoryObjectReceiveStream, MemoryObjectSendStream
from loguru import logger
from mcp import ClientSession, StdioServerParameters, Tool as MCPTool, stdio_client
from mcp.client.sse import sse_client
from mcp.types import CallToolResult, JSONRPCMessage
from swarms.utils.any_to_str import any_to_str
class MCPServer(abc.ABC):
"""Base class for Model Context Protocol servers."""
@abc.abstractmethod
async def connect(self) -> None:
"""Establish connection to the MCP server."""
pass
@property
@abc.abstractmethod
def name(self) -> str:
"""Human-readable server name."""
pass
@abc.abstractmethod
async def cleanup(self) -> None:
"""Clean up resources and close connection."""
pass
@abc.abstractmethod
async def list_tools(self) -> List[MCPTool]:
"""List available MCP tools on the server."""
pass
@abc.abstractmethod
async def call_tool(
self, tool_name: str, arguments: Dict[str, Any] | None
) -> CallToolResult:
"""Invoke a tool by name with provided arguments."""
pass
class _MCPServerWithClientSession(MCPServer, abc.ABC):
"""Mixin providing ClientSession-based MCP communication."""
def __init__(self, cache_tools_list: bool = False):
self.session: Optional[ClientSession] = None
self.exit_stack: AsyncExitStack = AsyncExitStack()
self._cleanup_lock = asyncio.Lock()
self.cache_tools_list = cache_tools_list
self._cache_dirty = True
self._tools_list: Optional[List[MCPTool]] = None
@abc.abstractmethod
def create_streams(
self
) -> AbstractAsyncContextManager[
tuple[
MemoryObjectReceiveStream[JSONRPCMessage | Exception],
MemoryObjectSendStream[JSONRPCMessage],
]
]:
"""Supply the read/write streams for the MCP transport."""
pass
async def __aenter__(self) -> MCPServer:
await self.connect()
return self # type: ignore
async def __aexit__(self, exc_type, exc_value, tb) -> None:
await self.cleanup()
async def connect(self) -> None:
"""Initialize transport and ClientSession."""
try:
transport = await self.exit_stack.enter_async_context(
self.create_streams()
)
read, write = transport
session = await self.exit_stack.enter_async_context(
ClientSession(read, write)
)
await session.initialize()
self.session = session
except Exception as e:
logger.error(f"Error initializing MCP server: {e}")
await self.cleanup()
raise
async def cleanup(self) -> None:
"""Close session and transport."""
async with self._cleanup_lock:
try:
await self.exit_stack.aclose()
except Exception as e:
logger.error(f"Error during cleanup: {e}")
finally:
self.session = None
async def list_tools(self) -> List[MCPTool]:
if not self.session:
raise RuntimeError("Server not connected. Call connect() first.")
if self.cache_tools_list and not self._cache_dirty and self._tools_list:
return self._tools_list
self._cache_dirty = False
self._tools_list = (await self.session.list_tools()).tools
return self._tools_list # type: ignore
async def call_tool(
self, tool_name: str | None = None, arguments: Dict[str, Any] | None = None
) -> CallToolResult:
if not arguments:
raise ValueError("Arguments dict is required to call a tool")
name = tool_name or arguments.get("tool_name") or arguments.get("name")
if not name:
raise ValueError("Tool name missing in arguments")
if not self.session:
raise RuntimeError("Server not connected. Call connect() first.")
return await self.session.call_tool(name, arguments)
class MCPServerStdioParams(TypedDict):
"""Configuration for stdio transport."""
command: str
args: NotRequired[List[str]]
env: NotRequired[Dict[str, str]]
cwd: NotRequired[str | Path]
encoding: NotRequired[str]
encoding_error_handler: NotRequired[Literal["strict", "ignore", "replace"]]
class MCPServerStdio(_MCPServerWithClientSession):
"""MCP server over stdio transport."""
def __init__(
self,
params: MCPServerStdioParams,
cache_tools_list: bool = False,
name: Optional[str] = None,
):
super().__init__(cache_tools_list)
self.params = StdioServerParameters(
command=params["command"],
args=params.get("args", []),
env=params.get("env"),
cwd=params.get("cwd"),
encoding=params.get("encoding", "utf-8"),
encoding_error_handler=params.get("encoding_error_handler", "strict"),
)
self._name = name or f"stdio:{self.params.command}"
def create_streams(self) -> AbstractAsyncContextManager[
tuple[
MemoryObjectReceiveStream[JSONRPCMessage | Exception],
MemoryObjectSendStream[JSONRPCMessage],
]
]:
return stdio_client(self.params)
@property
def name(self) -> str:
return self._name
class MCPServerSseParams(TypedDict):
"""Configuration for HTTP+SSE transport."""
url: str
headers: NotRequired[Dict[str, str]]
timeout: NotRequired[float]
sse_read_timeout: NotRequired[float]
class MCPServerSse(_MCPServerWithClientSession):
"""MCP server over HTTP with SSE transport."""
def __init__(
self,
params: MCPServerSseParams,
cache_tools_list: bool = False,
name: Optional[str] = None,
):
super().__init__(cache_tools_list)
self.params = params
self._name = name or f"sse:{params['url']}"
def create_streams(self) -> AbstractAsyncContextManager[
tuple[
MemoryObjectReceiveStream[JSONRPCMessage | Exception],
MemoryObjectSendStream[JSONRPCMessage],
]
]:
return sse_client(
url=self.params["url"],
headers=self.params.get("headers"),
timeout=self.params.get("timeout", 5),
sse_read_timeout=self.params.get("sse_read_timeout", 300),
)
@property
def name(self) -> str:
return self._name
async def call_tool_fast(
server: MCPServerSse, payload: Dict[str, Any] | str
) -> Any:
"""Async function to call a tool on a server with proper cleanup."""
try:
await server.connect()
arguments = payload if isinstance(payload, dict) else None
result = await server.call_tool(arguments=arguments)
return result
finally:
await server.cleanup()
async def mcp_flow_get_tool_schema(
params: MCPServerSseParams,
) -> Any:
"""Async function to get tool schema from MCP server."""
async with MCPServerSse(params) as server:
tools = await server.list_tools()
return tools
async def mcp_flow(
params: MCPServerSseParams,
function_call: Dict[str, Any] | str,
) -> Any:
"""Async function to call a tool with given parameters."""
async with MCPServerSse(params) as server:
return await call_tool_fast(server, function_call)
async def _call_one_server(
params: MCPServerSseParams, payload: Dict[str, Any] | str
) -> Any:
"""Helper function to call a single MCP server."""
server = MCPServerSse(params)
try:
await server.connect()
arguments = payload if isinstance(payload, dict) else None
return await server.call_tool(arguments=arguments)
finally:
await server.cleanup()
async def abatch_mcp_flow(
params: List[MCPServerSseParams], payload: Dict[str, Any] | str
) -> List[Any]:
"""Async function to execute a batch of MCP calls concurrently.
Args:
params (List[MCPServerSseParams]): List of MCP server configurations
payload (Dict[str, Any] | str): The payload to send to each server
Returns:
List[Any]: Results from all MCP servers
"""
if not params:
logger.warning("No MCP servers provided for batch operation")
return []
try:
return await asyncio.gather(*[_call_one_server(p, payload) for p in params])
except Exception as e:
logger.error(f"Error in abatch_mcp_flow: {e}")
# Return partial results if any were successful
return [f"Error in batch operation: {str(e)}"]
def batch_mcp_flow(
params: List[MCPServerSseParams], payload: Dict[str, Any] | str
) -> List[Any]:
"""Sync wrapper for batch MCP operations.
This creates a new event loop if needed to run the async batch operation.
ONLY use this when not already in an async context.
Args:
params (List[MCPServerSseParams]): List of MCP server configurations
payload (Dict[str, Any] | str): The payload to send to each server
Returns:
List[Any]: Results from all MCP servers
"""
if not params:
logger.warning("No MCP servers provided for batch operation")
return []
try:
# Check if we're already in an event loop
try:
loop = asyncio.get_event_loop()
except RuntimeError:
# No event loop exists, create one
loop = asyncio.new_event_loop()
asyncio.set_event_loop(loop)
if loop.is_running():
# We're already in an async context, can't use asyncio.run
# Use a future to bridge sync-async gap
future = asyncio.run_coroutine_threadsafe(
abatch_mcp_flow(params, payload), loop
)
return future.result(timeout=30) # Add timeout to prevent hanging
else:
# We're not in an async context, safe to use loop.run_until_complete
return loop.run_until_complete(abatch_mcp_flow(params, payload))
except Exception as e:
logger.error(f"Error in batch_mcp_flow: {e}")
return [f"Error in batch operation: {str(e)}"]
from __future__ import annotations
import abc
import asyncio
from contextlib import AbstractAsyncContextManager, AsyncExitStack
from pathlib import Path
from typing import Any, Dict, List, Optional, Literal, Union
from typing_extensions import NotRequired, TypedDict
from anyio.streams.memory import MemoryObjectReceiveStream, MemoryObjectSendStream
from loguru import logger
from mcp import ClientSession, StdioServerParameters, Tool as MCPTool, stdio_client
from mcp.client.sse import sse_client
from mcp.types import CallToolResult, JSONRPCMessage
from swarms.utils.any_to_str import any_to_str
class MCPServer(abc.ABC):
"""Base class for Model Context Protocol servers."""
@abc.abstractmethod
async def connect(self) -> None:
"""Establish connection to the MCP server."""
pass
@property
@abc.abstractmethod
def name(self) -> str:
"""Human-readable server name."""
pass
@abc.abstractmethod
async def cleanup(self) -> None:
"""Clean up resources and close connection."""
pass
@abc.abstractmethod
async def list_tools(self) -> List[MCPTool]:
"""List available MCP tools on the server."""
pass
@abc.abstractmethod
async def call_tool(self, tool_name: str,
arguments: Dict[str, Any] | None) -> CallToolResult:
"""Invoke a tool by name with provided arguments."""
pass
class _MCPServerWithClientSession(MCPServer, abc.ABC):
"""Mixin providing ClientSession-based MCP communication."""
def __init__(self, cache_tools_list: bool = False):
self.session: Optional[ClientSession] = None
self.exit_stack: AsyncExitStack = AsyncExitStack()
self._cleanup_lock = asyncio.Lock()
self.cache_tools_list = cache_tools_list
self._cache_dirty = True
self._tools_list: Optional[List[MCPTool]] = None
@abc.abstractmethod
def create_streams(
self
) -> AbstractAsyncContextManager[tuple[
MemoryObjectReceiveStream[JSONRPCMessage | Exception],
MemoryObjectSendStream[JSONRPCMessage],
]]:
"""Supply the read/write streams for the MCP transport."""
pass
async def __aenter__(self) -> MCPServer:
await self.connect()
return self # type: ignore
async def __aexit__(self, exc_type, exc_value, tb) -> None:
await self.cleanup()
async def connect(self) -> None:
"""Initialize transport and ClientSession."""
try:
transport = await self.exit_stack.enter_async_context(
self.create_streams())
read, write = transport
session = await self.exit_stack.enter_async_context(
ClientSession(read, write))
await session.initialize()
self.session = session
except Exception as e:
logger.error(f"Error initializing MCP server: {e}")
await self.cleanup()
raise
async def cleanup(self) -> None:
"""Close session and transport."""
async with self._cleanup_lock:
try:
await self.exit_stack.aclose()
except Exception as e:
logger.error(f"Error during cleanup: {e}")
finally:
self.session = None
async def list_tools(self) -> List[MCPTool]:
if not self.session:
raise RuntimeError("Server not connected. Call connect() first.")
if self.cache_tools_list and not self._cache_dirty and self._tools_list:
return self._tools_list
self._cache_dirty = False
self._tools_list = (await self.session.list_tools()).tools
return self._tools_list # type: ignore
async def call_tool(
self,
tool_name: str | None = None,
arguments: Dict[str, Any] | None = None) -> CallToolResult:
if not arguments:
raise ValueError("Arguments dict is required to call a tool")
name = tool_name or arguments.get("tool_name") or arguments.get("name")
if not name:
raise ValueError("Tool name missing in arguments")
if not self.session:
raise RuntimeError("Server not connected. Call connect() first.")
return await self.session.call_tool(name, arguments)
class MCPServerStdioParams(TypedDict):
"""Configuration for stdio transport."""
command: str
args: NotRequired[List[str]]
env: NotRequired[Dict[str, str]]
cwd: NotRequired[str | Path]
encoding: NotRequired[str]
encoding_error_handler: NotRequired[Literal["strict", "ignore", "replace"]]
class MCPServerStdio(_MCPServerWithClientSession):
"""MCP server over stdio transport."""
def __init__(
self,
params: MCPServerStdioParams,
cache_tools_list: bool = False,
name: Optional[str] = None,
):
super().__init__(cache_tools_list)
self.params = StdioServerParameters(
command=params["command"],
args=params.get("args", []),
env=params.get("env"),
cwd=params.get("cwd"),
encoding=params.get("encoding", "utf-8"),
encoding_error_handler=params.get("encoding_error_handler",
"strict"),
)
self._name = name or f"stdio:{self.params.command}"
def create_streams(
self
) -> AbstractAsyncContextManager[tuple[
MemoryObjectReceiveStream[JSONRPCMessage | Exception],
MemoryObjectSendStream[JSONRPCMessage],
]]:
return stdio_client(self.params)
@property
def name(self) -> str:
return self._name
class MCPServerSseParams(TypedDict):
"""Configuration for HTTP+SSE transport."""
url: str
headers: NotRequired[Dict[str, str]]
timeout: NotRequired[float]
sse_read_timeout: NotRequired[float]
class MCPServerSse(_MCPServerWithClientSession):
"""MCP server over HTTP with SSE transport."""
def __init__(
self,
params: MCPServerSseParams,
cache_tools_list: bool = False,
name: Optional[str] = None,
):
super().__init__(cache_tools_list)
self.params = params
self._name = name or f"sse:{params['url']}"
def create_streams(
self
) -> AbstractAsyncContextManager[tuple[
MemoryObjectReceiveStream[JSONRPCMessage | Exception],
MemoryObjectSendStream[JSONRPCMessage],
]]:
return sse_client(
url=self.params["url"],
headers=self.params.get("headers"),
timeout=self.params.get("timeout", 5),
sse_read_timeout=self.params.get("sse_read_timeout", 300),
)
@property
def name(self) -> str:
return self._name
async def call_tool_fast(server: MCPServerSse,
payload: Dict[str, Any] | str) -> Any:
"""Async function to call a tool on a server with proper cleanup."""
try:
await server.connect()
arguments = payload if isinstance(payload, dict) else None
result = await server.call_tool(arguments=arguments)
return result
finally:
await server.cleanup()
async def mcp_flow_get_tool_schema(params: MCPServerSseParams, ) -> Any:
"""Async function to get tool schema from MCP server."""
async with MCPServerSse(params) as server:
tools = await server.list_tools()
return tools
async def mcp_flow(
params: MCPServerSseParams,
function_call: Dict[str, Any] | str,
) -> Any:
"""Async function to call a tool with given parameters."""
async with MCPServerSse(params) as server:
return await call_tool_fast(server, function_call)
async def _call_one_server(params: MCPServerSseParams,
payload: Dict[str, Any] | str) -> Any:
"""Helper function to call a single MCP server."""
server = MCPServerSse(params)
try:
await server.connect()
arguments = payload if isinstance(payload, dict) else None
return await server.call_tool(arguments=arguments)
finally:
await server.cleanup()
async def abatch_mcp_flow(params: List[MCPServerSseParams],
payload: Dict[str, Any] | str) -> List[Any]:
"""Async function to execute a batch of MCP calls concurrently.
Args:
params (List[MCPServerSseParams]): List of MCP server configurations
payload (Dict[str, Any] | str): The payload to send to each server
Returns:
List[Any]: Results from all MCP servers
"""
if not params:
logger.warning("No MCP servers provided for batch operation")
return []
try:
return await asyncio.gather(
*[_call_one_server(p, payload) for p in params])
except Exception as e:
logger.error(f"Error in abatch_mcp_flow: {e}")
# Return partial results if any were successful
return [f"Error in batch operation: {str(e)}"]
def batch_mcp_flow(params: List[MCPServerSseParams],
payload: Dict[str, Any] | str) -> List[Any]:
"""Sync wrapper for batch MCP operations.
This creates a new event loop if needed to run the async batch operation.
ONLY use this when not already in an async context.
Args:
params (List[MCPServerSseParams]): List of MCP server configurations
payload (Dict[str, Any] | str): The payload to send to each server
Returns:
List[Any]: Results from all MCP servers
"""
if not params:
logger.warning("No MCP servers provided for batch operation")
return []
try:
# Check if we're already in an event loop
try:
loop = asyncio.get_event_loop()
except RuntimeError:
# No event loop exists, create one
loop = asyncio.new_event_loop()
asyncio.set_event_loop(loop)
if loop.is_running():
# We're already in an async context, can't use asyncio.run
# Use a future to bridge sync-async gap
future = asyncio.run_coroutine_threadsafe(
abatch_mcp_flow(params, payload), loop)
return future.result(timeout=30) # Add timeout to prevent hanging
else:
# We're not in an async context, safe to use loop.run_until_complete
return loop.run_until_complete(abatch_mcp_flow(params, payload))
except Exception as e:
logger.error(f"Error in batch_mcp_flow: {e}")
return [f"Error in batch operation: {str(e)}"]

Loading…
Cancel
Save