langgraph

LangGraph + odoo-mcp

Uses langchain-mcp-adapters (MultiServerMCPClient) to load the odoo-mcp tools and langchain.agents.create_agent to drive them.

Stale-tutorial warning: pin langchain-mcp-adapters >= 0.2.2. Versions 0.1.2/0.1.3 were yanked, and older create_react_agent examples no longer reflect the current API.

Run

pip install -r requirements.txt
export OPENAI_API_KEY=sk-...

# terminal 1 — the MCP server (set ODOO_* env vars first)
odoo-mcp --transport streamable-http --host 127.0.0.1 --port 8000 --path /mcp

# terminal 2
python odoo_agent.py

Expected output: a short list of draft sale orders (or an access diagnosis if the credential cannot read them).

stdio variant

Replace the server entry with:

"odoo": {
    "transport": "stdio",
    "command": "uvx",
    "args": ["odoo-mcp"],
    "env": {
        "ODOO_URL": "...", "ODOO_DB": "...",
        "ODOO_USERNAME": "...", "ODOO_PASSWORD": "...",
    },
},

Anti-pattern: do not use stdio when deploying on LangGraph Server / LangGraph Cloud — the platform may spawn multiple workers, each forking its own odoo-mcp subprocess. Run one Streamable HTTP server instead.

Multi-instance

With a multi-instance odoo_config.json, mention the instance in the prompt ("using the client_a instance, …") — every odoo-mcp tool accepts an instance argument the agent will fill in.

---

Last verified: 2026-06-10 — ran end-to-end against langchain==1.3.6, langchain-mcp-adapters==0.2.2, langgraph==1.2.4, odoo-mcp Streamable HTTP, a live Odoo 19 (Docker), and DeepSeek (OPENAI_BASE_URL + OPENAI_MODEL=deepseek-chat).