MCP · Model Context Protocol

Every Ollagraph endpoint,
a tool your agent can call.

One install, one env var, and Claude Desktop / Cursor / Cline / Continue gain every Ollagraph endpoint as a callable tool. Built on FastMCP, auto-generated from our OpenAPI spec, so every new endpoint appears the moment we ship it.

Get an API key GitHub repo →

INSTALL

Run the server.

Two equivalent install paths. Both fetch ollagraph-mcp from PyPI.

# Option A: ephemeral (no install)
export OLLAGRAPH_API_KEY="osk_..."
pipx run ollagraph-mcp

# Option B: persistent
pip install ollagraph-mcp
ollagraph-mcp

CLAUDE DESKTOP

Drop into your config.

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "ollagraph": {
      "command": "pipx",
      "args": ["run", "ollagraph-mcp"],
      "env": {
        "OLLAGRAPH_API_KEY": "osk_..."
      }
    }
  }
}

Restart Claude Desktop. The tools (scrape, extract_clean, aeo_page_audit, …) appear in the tool picker.

OTHER CLIENTS

Cursor, Cline, Continue.

Same JSON shape. Each client has its own config location — see their MCP docs. To poke the tools directly without a client, open the MCP Inspector:

npx @modelcontextprotocol/inspector pipx run ollagraph-mcp

WHAT YOU GET

Every endpoint, as a tool.

Auto-synced with the API

The server reads api.ollagraph.com/openapi.json at start. Restart the server when we ship a new endpoint — your agent gets the new tool.

Bearer auth handled

Set OLLAGRAPH_API_KEY once. The server injects the header on every call — the LLM never sees your key.

Honest tool descriptions

Each tool's description is its OpenAPI summary. The LLM reads them to pick which to call — no manual prompt engineering.

MIT-licensed, open source

The MCP server source is on GitHub. Set the API base via OLLAGRAPH_API_URL.

Get a key, run the server.

1,000 free credits on signup. No card required.

Start free All API endpoints

Every Ollagraph endpoint, a tool your agent can call.