Dawo MCP Server

Use Dawo's investment-research tools from any MCP host (Claude Desktop, Cursor, …) — pull a Lead Analyst verdict, run a research plan, screen, compare, stress-test, and more, right inside your AI.

It's a thin proxy: every tool runs on Dawo's backend (identical to the in-app advisor). Nothing is computed locally. It's a Pro perk — you authenticate with your own Dawo API key, and tool execution stays Pro-gated.

1. Mint a Dawo API key (Pro)

In the Dawo web app: Settings → API keys → Create key. Copy the dwo_… key — it's shown once. (Or via API: POST /auth/api-keys with your logged-in session.)

Treat the key like a password. You can revoke it anytime (the key stops working immediately, and only works while your account is Pro).

2. Add it to your MCP host

Claude Desktop — edit claude_desktop_config.json:

{
  "mcpServers": {
    "dawo": {
      "command": "uvx",
      "args": ["dawo-mcp-server"],
      "env": {
        "DAWO_API_KEY": "dwo_xxxxxxxxxxxxxxxxxxxxxxxxxxxx",
        "DAWO_BASE_URL": "https://api-morning-silence-6501.fly.dev"
      }
    }
  }
}

DAWO_BASE_URL is optional (defaults to production). Restart the host and ask, e.g., "Use Dawo to give me the verdict on NVDA" or "With Dawo, build a research plan for quality growth under $50B."

3. Run locally (dev)

cd services/mcp-server
pip install -e .
DAWO_API_KEY=dwo_... dawo-mcp-server   # speaks MCP over stdio

Or without installing: DAWO_API_KEY=dwo_... uvx --from . dawo-mcp-server.

How it works

• On startup it calls GET /engines/tools/list and registers each tool (name + JSON schema) as an MCP tool.
• On a tool call it POSTs to /engines/tools/execute with {tool, params} and your key.
• 401/403 → your key is invalid/revoked or your plan isn't Pro; 503 → tools API temporarily off.

No data, secrets, or logic are stored locally beyond the API key you provide via env.

Publishing (maintainers)

Publishing is automated by the publish-mcp GitHub Action. Bump version in pyproject.toml, then push a matching tag — the workflow builds the wheel + sdist and uploads to PyPI:

# after bumping version in pyproject.toml (e.g. to 0.1.0)
git tag mcp-v0.1.0 && git push origin mcp-v0.1.0

It authenticates with the PYPI_API_TOKEN repo secret (Settings → Secrets and variables → Actions). You can also trigger it manually from the Actions tab. To build/upload locally instead:

cd services/mcp-server
python -m build && python -m twine upload dist/*   # TWINE_PASSWORD=pypi-…

The canonical install is the PyPI package — uvx dawo-mcp-server (see §2) — once published. The uvx --from git+https://github.com/…#subdirectory=services/mcp-server form only works for callers who can access the repo, so it's a dev convenience, not the public install path.