Code-Fundi MCP Server

A production-grade Model Context Protocol (MCP) server for the Code-Fundi API: a codebase map and blast-radius engine for AI coding agents. Enables any MCP-compatible AI assistant (Claude, Claude Code, Cursor, Windsurf, VS Code Copilot, and more) to search, research, index, and scan code repositories through natural language, and see what a change breaks before it ships.

Built with FastMCP (TypeScript) and Zod for schema validation.

Getting started for AI agents

If you are an AI agent (or configuring one) and no CODEFUNDI_API_KEY is set, you must self-authenticate before using search, repos, chat, or other authenticated tools. The auth tools (code-fundi-auth-*) work without a pre-existing key.

Which path to use

| Situation | Action | |-----------|--------| | CODEFUNDI_API_KEY is set in MCP config | Use any tool immediately — the server sends it as X-API-Key on every request | | No API key configured | Run the OTP self-auth flow below (recommended) or password sign-in |

Zero-config MCP setup (no API key)

You can start the MCP server without CODEFUNDI_API_KEY and authenticate at runtime:

{
  "mcpServers": {
    "code-fundi": {
      "command": "npx",
      "args": ["-y", "@codefundi/code-fundi-mcp"]
    }
  }
}

Self-authenticate with OTP (recommended)

Code-Fundi uses Supabase-backed auth (POST /v2/auth/authenticate, /v2/auth/verify, /v2/auth/resend). OTP emails contain a 6-digit code — magic links are not supported on this path.

1. Ask the human user for their email address.
2. Call code-fundi-auth-authenticate with:
   • auth_mode: "otp"
   • email: the user's email
   • should_create_user: true for a new account, false for a returning user
3. Tell the user to check their inbox for a 6-digit code. The API may return verification_required: true and api_key.key_state: "agent_pending" until verification completes.
4. Ask the user for the 6-digit OTP (human-in-the-loop — you cannot guess or bypass this step).
5. Call code-fundi-auth-verify with the same email and the token (6 digits).
6. On success, the MCP server automatically configures the API key in memory for all subsequent tool calls in this session.

If the code expired or was not received, call code-fundi-auth-resend with the same email, then repeat step 5.

Example dialogue:

Agent: What email should I use to sign in to Code-Fundi?
User:  [email protected]
Agent: [calls code-fundi-auth-authenticate] I've sent a 6-digit code to [email protected]. Please paste it here.
User:  482913
Agent: [calls code-fundi-auth-verify] You're signed in. I can now search and index your repositories.

Password sign-in (alternative)

For existing accounts with a password, call code-fundi-auth-authenticate with auth_mode: "password", the user's email, should_create_user: false, and the password parameter. The MCP client sends the password only in the X-CodeFundi-Auth-Password header (never in the JSON body). Production requires HTTPS. Returning users may receive an active API key immediately without a separate verify step.

After authentication

• The API key is held in memory for the lifetime of the MCP server process. It is not persisted across IDE or MCP restarts.
• Recommend the user add the key to their MCP config as CODEFUNDI_API_KEY so future sessions start authenticated.
• A FREE-tier account and API key are created automatically on first signup.

Errors

| HTTP status | What to do | |-------------|------------| | 401 Unauthorized | No valid key — run the OTP flow above, or set CODEFUNDI_API_KEY | | 429 Too many requests | Auth endpoints are rate-limited per IP; wait for Retry-After seconds, then retry |

Features

Every tool below is backed by the same codebase map: structural dependencies, call graph, and blast radius, indexed once and queried in milliseconds.

• 🔍 Semantic & grep code search across your indexed codebase map
• 🧠 AI-powered research: search plus AI analysis in a single call
• 📦 Repository management: index, status, README, listing, public catalog
• 🛰️ Repository intelligence: cross-repo dependency map, blueprint, Blast-Radius Guard (impact analysis before you merge)
• 📄 File documentation: AI-generated docs for any indexed file
• 📊 Usage statistics: query usage, activity, language breakdowns
• 🔐 Agent-driven authentication: sign up/sign in via OTP without pre-configured keys
• 💬 AI chat & model insight: direct conversation with Code-Fundi AI, model catalog, and per-tier limits

Quick Start

Install from npm (recommended)

Install the package (includes a pre-built dist/). The code-fundi-mcp binary is on your PATH when installed globally, or available via npx without cloning the repo:

npm install -g @codefundi/code-fundi-mcp

Or add it to a project:

npm install @codefundi/code-fundi-mcp

Install from source (this repository)

git clone https://github.com/Code-Fundi/code-fundi-mcp.git
cd code-fundi-mcp
npm install
npm run build

Configure

Option A — API key (fastest): set your key as an environment variable:

export CODEFUNDI_API_KEY=your_api_key_here

Option B — no API key: skip the env var and let the agent self-authenticate at runtime. See Getting started for AI agents.

Zero-config MCP example (no env block):

{
  "mcpServers": {
    "code-fundi": {
      "command": "npx",
      "args": ["-y", "@codefundi/code-fundi-mcp"]
    }
  }
}

Use with Claude Desktop

After a global install (npm i -g @codefundi/code-fundi-mcp), point MCP at the published binary (no path to dist/index.js required):

{
  "mcpServers": {
    "code-fundi": {
      "command": "@codefundi/code-fundi-mcp",
      "env": {
        "CODEFUNDI_API_KEY": "your_api_key_here"
      }
    }
  }
}

If the binary is not on your PATH, use npx (downloads or uses the local package and runs the same entrypoint):

{
  "mcpServers": {
    "code-fundi": {
      "command": "npx",
      "args": ["-y", "@codefundi/code-fundi-mcp"],
      "env": {
        "CODEFUNDI_API_KEY": "your_api_key_here"
      }
    }
  }
}

Use with Cursor

Same pattern as Claude: command plus optional args only, no manual path to the repo:

{
  "mcpServers": {
    "code-fundi": {
      "command": "npx",
      "args": ["-y", "@codefundi/code-fundi-mcp"],
      "env": {
        "CODEFUNDI_API_KEY": "your_api_key_here"
      }
    }
  }
}

When developing inside this repo, you can run npm run dev or npx tsx src/index.ts without building first.

Development Mode

npm run dev                          # Run with tsx (no build needed)
npx fastmcp inspect src/index.ts    # Open MCP Inspector UI
npx fastmcp dev src/index.ts        # Test with MCP CLI

Tools Reference (27 tools)

Covers the Code-Fundi V2 API for codebase mapping and blast-radius analysis: search (including search-with-chat / research), repositories (list, index, status, readme, public catalog), repository intelligence (map, blueprint, radius), files, history, statistics, API keys, authentication, plus Fundi chat (POST /v1/fundi/chat) and the V2 model catalog / limits (GET /v2/models, GET /v2/models/limits).

Search

| Tool | Description | |------|-------------| | code-fundi-search | Semantic and grep search across your indexed codebase map, with filters | | code-fundi-research | Search plus AI-synthesized analysis of matching code |

Repositories

| Tool | Description | |------|-------------| | code-fundi-list-repos | List indexed repositories with pagination | | code-fundi-index-repo | Index a new GitHub repository into your codebase map | | code-fundi-repo-status | Check repository indexing status | | code-fundi-repo-readme | Get repository README documentation (deprecated, prefer blueprint) | | code-fundi-list-public-repos | Browse the global catalog of indexed public repositories (no key needed) |

Repository Intelligence

| Tool | Description | |------|-------------| | code-fundi-repo-map | Cross-repository dependency map: how services and packages actually connect | | code-fundi-repo-blueprint | README plus dependency and convention overview (successor to repo-readme) | | code-fundi-repo-radius | Blast-Radius Guard: every file and function that breaks before you merge (PRO+) |

Files

| Tool | Description | |------|-------------| | code-fundi-list-files | List files in a repository | | code-fundi-file-docs | Get AI-generated file documentation |

History

| Tool | Description | |------|-------------| | code-fundi-list-history | List query history with filters | | code-fundi-history-item | Get full details of a history entry | | code-fundi-conversation | Get conversation thread messages |

Statistics

| Tool | Description | |------|-------------| | code-fundi-usage-stats | Query-type usage breakdown | | code-fundi-activity-stats | Daily activity statistics | | code-fundi-language-stats | Programming language usage |

Authentication

| Tool | Description | |------|-------------| | code-fundi-auth-authenticate | Start OTP/password auth flow (no key needed) | | code-fundi-auth-verify | Verify OTP code and obtain API key | | code-fundi-auth-resend | Resend OTP verification email | | code-fundi-list-api-keys | List API keys (masked) | | code-fundi-regenerate-api-key | Regenerate API key | | code-fundi-disable-api-key | Disable an API key by ID (DELETE /v2/keys/{key_id}) |

Chat & Models

| Tool | Description | |------|-------------| | code-fundi-chat | Fundi AI chat (POST /v1/fundi/chat; streamed responses are collected to text) | | code-fundi-list-models | List the curated chat model catalog (GET /v2/models) | | code-fundi-model-limits | Get AI model limits and tier configuration (GET /v2/models/limits) |

Authentication

Two modes: pre-configured API key (CODEFUNDI_API_KEY in MCP config) or agent-driven OTP/password auth at runtime. Full step-by-step instructions, tool names, and error handling are in Getting started for AI agents at the top of this README.

Environment Variables

| Variable | Required | Default | Description | |----------|----------|---------|-------------| | CODEFUNDI_API_KEY | No* | — | API key for authentication | | CODEFUNDI_BASE_URL | No | https://api.codefundi.app | API base URL override |

* Required unless using agent-driven auth tools.

License

MIT