|
|
### Why you’re seeing **`name 'batch_mcp_flow' is not defined`**
|
|
|
|
|
|
1. In **`swarms/structs/agent.py`** the method `mcp_execution_flow()` (or the block that calls MCP) still tries to invoke `batch_mcp_flow`.
|
|
|
2. At the very top of the file the import that *defines* that symbol was commented out when you switched to fast‑MCP:
|
|
|
|
|
|
```python
|
|
|
# from swarms.tools.mcp_integration import (
|
|
|
# MCPServerSseParams,
|
|
|
# batch_mcp_flow, # ← you lost this
|
|
|
# mcp_flow_get_tool_schema,
|
|
|
# )
|
|
|
```
|
|
|
|
|
|
3. When Python hits `batch_mcp_flow(…)` it isn’t in the module’s namespace, so the call blows up and your agent prints the MCP‑error.
|
|
|
|
|
|
---
|
|
|
|
|
|
## Two‑step fix
|
|
|
|
|
|
> Below assumes your `swarms/tools/mcp_integration.py` already contains a working (sync) wrapper named `batch_mcp_flow` that *returns a string*, **not** a coroutine object.
|
|
|
> If you kept it **async**, see the “If your wrapper is still async” note at the end.
|
|
|
|
|
|
---
|
|
|
|
|
|
### 1 · Restore the import in **`agent.py`**
|
|
|
|
|
|
Un‑comment (or add) the line:
|
|
|
|
|
|
```python
|
|
|
from swarms.tools.mcp_integration import batch_mcp_flow
|
|
|
```
|
|
|
|
|
|
Keep the `MCPServerSseParams` import too if you reference it elsewhere.
|
|
|
|
|
|
---
|
|
|
|
|
|
### 2 · Make sure `batch_mcp_flow()` is **synchronous**
|
|
|
|
|
|
Inside **`swarms/tools/mcp_integration.py`** do something like:
|
|
|
|
|
|
```python
|
|
|
# ── swarms/tools/mcp_integration.py ─────────────────────
|
|
|
import asyncio
|
|
|
# … other imports …
|
|
|
|
|
|
async def _batch_mcp_flow_async(
|
|
|
params: list[MCPServerSseParams],
|
|
|
function_call: dict[str, Any],
|
|
|
) -> str:
|
|
|
# your existing async logic that talks to MCP
|
|
|
# and returns a JSON‑serialisable string/str
|
|
|
...
|
|
|
|
|
|
def batch_mcp_flow(
|
|
|
params: list[MCPServerSseParams],
|
|
|
function_call: dict[str, Any],
|
|
|
) -> str:
|
|
|
"""Blocking wrapper so callers don't have to await."""
|
|
|
return asyncio.run(
|
|
|
_batch_mcp_flow_async(params, function_call)
|
|
|
)
|
|
|
```
|
|
|
|
|
|
Now `agent.py` can call `batch_mcp_flow(...)` directly and
|
|
|
get **the tool’s real answer**, not a coroutine object.
|
|
|
|
|
|
---
|
|
|
|
|
|
### Quick checklist
|
|
|
|
|
|
- [ ] Re‑import `batch_mcp_flow` in `swarms/structs/agent.py`.
|
|
|
- [ ] Ensure `batch_mcp_flow()` (the public one) **returns a value**, not a coroutine.
|
|
|
- [ ] Remove or comment out the temporary warning prints you added earlier.
|
|
|
|
|
|
---
|
|
|
|
|
|
### If your wrapper is still async
|
|
|
|
|
|
You can alternatively **await** it inside `mcp_execution_flow`:
|
|
|
|
|
|
```python
|
|
|
import asyncio
|
|
|
|
|
|
def mcp_execution_flow(self, response: str) -> str:
|
|
|
try:
|
|
|
resp_dict = json.loads(response)
|
|
|
return asyncio.run(
|
|
|
batch_mcp_flow(self.mcp_servers, resp_dict)
|
|
|
)
|
|
|
except Exception as e:
|
|
|
logger.error(f"MCP flow failed: {e}")
|
|
|
return f"[MCP-error] {e}"
|
|
|
```
|
|
|
|
|
|
Either approach works; the key is that the call you make inside the agent must not leave a bare coroutine object floating around.
|
|
|
|
|
|
Once those two lines are fixed, run the client again:
|
|
|
|
|
|
```bash
|
|
|
python examples/mcp_example/mcp_client.py
|
|
|
```
|
|
|
|
|
|
You should now see a clean answer:
|
|
|
|
|
|
```
|
|
|
Enter your query (or 'exit' to quit): add 3 and 33
|
|
|
Math Agent Response: 36
|
|
|
```
|
|
|
|
|
|
Happy calculating! |
