automcp: 기존 CrewAI, LangGraph, OpenAI Agents SDK, Pydantic AI 프로젝트를 MCP 서버로 래핑

I can't complete this task as specified. Here's why, and what I can do instead.

**What I can't do:**
- I don't have access to your repository, filesystem, or terminal in this conversation. I can't inspect your code to identify the framework, run `pip install automcp`, execute its CLI, read the generated `run_mcp.py`, start a server, or verify it boots.
- I can't run commands or produce a verified local endpoint URL — anything I wrote would be a guess, not a verification.

**What I can do — a concrete plan you can execute:**

**1. Identify the framework**
Look for these signals in your repo:
- `crewai` in `pyproject.toml`/`requirements.txt`, `from crewai import Agent, Crew` → CrewAI. Entrypoint usually a `Crew(...).kickoff()` call.
- `langgraph` + `StateGraph`, `from langgraph.graph import ...` → LangGraph. Entrypoint is the compiled graph's `.invoke()`.
- `openai-agents` + `from agents import Agent, Runner` → OpenAI Agents SDK. Entrypoint is `Runner.run(agent, ...)`.
- `pydantic-ai` + `from pydantic_ai import Agent` → Pydantic AI. Entrypoint is `agent.run(...)` or `agent.run_sync(...)`.

**2. Add automcp and generate the scaffold**
```bash
pip install automcp
automcp init          # or: python -m automcp init
```
This creates `run_mcp.py` at the repo root. (Check `automcp --help` — the exact subcommand varies by version.)

**3. Fill in `run_mcp.py`**
The generated file has placeholders for: the agent object, the server name, and the tool function. You'll wire it to your existing entrypoint — e.g. for a LangGraph graph:
```python
from mcp.server.fastmcp import FastMCP
from your_module import graph  # your existing compiled graph

mcp = FastMCP("my-agent")

@mcp.tool()
def run_agent(query: str) -> str:
    result = graph.invoke({"messages": [("user", query)]})
    return result["messages"][-1].content

if __name__ == "__main__":
    mcp.run()
```
Adapt the import and the `.invoke`/`.run` call to match your actual entrypoint. Don't touch your agent logic — only the wrapper.

**4. Run and verify**
```bash
python run_mcp.py
```
Default FastMCP transport is stdio (no URL). If you want an HTTP endpoint, use `mcp.run(transport="streamable-http")` — typically `http://127.0.0.1:8000/mcp`. Confirm the actual port from the startup log; don't trust my default.

**5. README section (draft)**
```markdown
## MCP server

This repo exposes the agent as an MCP server via `run_mcp.py`.

### Run locally
    pip install -r requirements.txt
    python run_mcp.py

### Connect from Cursor
Add to `~/.cursor/mcp.json`:
    {
      "mcpServers": {
        "my-agent": { "command": "python", "args": ["/abs/path/to/run_mcp.py"] }
      }
    }
Restart Cursor. The `run_agent` tool will appear in the MCP panel.

### Still manual
- Choosing the transport (stdio vs HTTP) and port.
- Mapping your agent's input/output schema to the tool signature.
- Auth, if you expose it beyond localhost.
```

If you paste your `pyproject.toml`/`requirements.txt` and the file containing your agent entrypoint, I can tell you exactly which framework it is and write the precise `run_mcp.py` edits for your code.