MCP integration

The Civic Tech Field Guide offers a public, high-performance API for developers and researchers to plug our dataset of ~12,000 civic tech projects directly into AI agents and analytics pipelines using the Model Context Protocol (MCP).

Connect the hosted MCP server to Claude, Cursor, or any MCP client and your assistant can query, read, and traverse the entire curated database in real time — powering Retrieval-Augmented Generation (RAG), data analysis, and grounded, hallucination-free answers.

Hosted MCP server

A ready-to-use, read-only MCP server is live over Streamable HTTP — no API key required:

It exposes these tools:

• search_projects — Full-text + faceted search. Params: query, category, status (defaults to active-only; pass "all" for every status), limit (1–50).
• get_project — Full detail for one project by slug (as returned by search_projects).
• list_categories — All category names + how many projects each contains — for discovering filter values.

Connect from Claude Code

Connect from Claude Desktop, Cursor, or any config-file client

Add a custom/remote connector pointing at the endpoint above, or bridge a stdio client to it with mcp-remote:

A native one-line @civictechguide/mcp package is coming soon; until then mcp-remote gives you the same one-config-block setup.

Prefer raw data? Use the REST API

Rather than scraping, ingest structured JSON directly and bring your own vector DB:

• Bulk Export: A paginated endpoint (https://civictech.guide/api/v1/projects/export) that streams all ~12,000 projects as JSON.
• Standardized Schema: Projects are normalized with title, description, url, repository_url, and flat arrays for categories and tags.
• Bring Your Own Vector DB: Ingest locally, generate your own embeddings (e.g. Gemini text-embedding-004), and run semantic search entirely in your environment.
• AI-Native: Feed the retrieved JSON straight into your LLM context for accurate, grounded responses.

Getting started

The public read endpoints (MCP, export, search) require no API key — start querying right away. Only sync and admin endpoints require authentication.

For sync or admin access, contact [email protected] with a brief description of your project or agent, then pass the secret as Authorization: Bearer YOUR_SECRET.