Procore MCP Server

MCP server that exposes the full Procore REST API to AI assistants like Claude. Built with TypeScript and the Model Context Protocol SDK.

Works with Claude Desktop, Claude Code, and any MCP-compatible client.

What it does

A build-time parser converts Procore's OpenAPI spec into a compact catalog, then auto-generates individual MCP tools for every API operation. At runtime, 7 meta-tools let the AI discover and call any Procore endpoint:

Prerequisites

• Node.js 18+
• A Procore Developer Portal account
• An OAuth app with Authorization Code grant type
• Set your redirect URI to http://localhost

Setup

git clone https://github.com/TylerIlunga/procore-mcp-server.git
cd procore-mcp-server
npm install

Copy the example env file and fill in your credentials:

cp .env.example .env

PROCORE_CLIENT_ID=your_client_id
PROCORE_CLIENT_SECRET=your_client_secret
PROCORE_COMPANY_ID=your_company_id

You'll need Procore's OpenAPI spec file placed at specs/combined_OAS.json. This file is not included in the repo due to its size (~54MB). You can obtain it from Procore's API documentation.

Build the catalog and compile TypeScript:

npm run build

Authenticate with Procore (opens browser for OAuth):

npm run auth

Start the server:

npm start

Claude Desktop configuration

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "procore": {
      "command": "node",
      "args": ["/absolute/path/to/procore-mcp-server/dist/src/index.js"],
      "env": {
        "PROCORE_CLIENT_ID": "your_client_id",
        "PROCORE_CLIENT_SECRET": "your_client_secret",
        "PROCORE_COMPANY_ID": "your_company_id"
      }
    }
  }
}

Claude Code configuration

Add to .mcp.json in your project root:

{
  "mcpServers": {
    "procore": {
      "command": "node",
      "args": ["/absolute/path/to/procore-mcp-server/dist/src/index.js"],
      "env": {
        "PROCORE_CLIENT_ID": "your_client_id",
        "PROCORE_CLIENT_SECRET": "your_client_secret",
        "PROCORE_COMPANY_ID": "your_company_id"
      }
    }
  }
}

Project structure

src/
  auth/       OAuth token exchange, refresh, storage
  api/        HTTP client with auth, rate limits, retries
  catalog/    Endpoint catalog loading, search, filtering
  tools/      MCP tool handlers and registration
scripts/
  generate-catalog.ts          Parse OAS into catalog
  generate-tools-manifest.ts   Generate per-endpoint MCP tools
  validate-catalog.ts          Validate catalog integrity
data/         Build output (committed): catalog.json, endpoint details, tools manifest
specs/        Source OAS file (gitignored — too large for the repo)

How it works

1. Build time: scripts/generate-catalog.ts parses the ~54MB Procore OpenAPI spec (3,155 operations) and produces a compact data/catalog.json plus individual endpoint detail files in data/endpoint-details/. scripts/generate-tools-manifest.ts then generates a tools manifest with one named MCP tool per API operation — 2,929 in total, after collapsing older-version duplicates of the same path.

   Each generated tool carries a structured description covering what it acts on, when to reach for it, which parent ids to resolve first, what it returns, and how it fails. Endpoints Procore has deprecated are registered with their sunset date in the description and a (Deprecated) title. The interactive /oauth/* endpoints are not registered as tools — npm run auth owns that flow — but remain reachable through procore_api_call.

2. Auth: Run npm run auth once to complete the OAuth flow in your browser. Tokens are saved to ~/.procore-mcp/tokens.json and auto-refresh when expired.

3. Runtime: The MCP server loads the catalog and registers all tools. When an AI assistant calls a tool, the server maps it to the correct Procore API endpoint, injects auth headers, handles rate limits and pagination, and returns the response.

Inspiration

Built to help my girlfriend, a construction engineer who uses Procore daily.

License

MIT