coldfusion-mcp

MCP server for semantic and structural analysis of ColdFusion (CFML) codebases. Provides 8 tools for searching, tracing dependencies, explaining components, and finding SQL usage across multi-application CF projects.

Built on Model Context Protocol for use with Claude Code, Claude Desktop, and other MCP-compatible clients.

Prerequisites

• Node.js >= 20.6.0
• Redis running locally (for graph caching)
• OpenAI API key (for embeddings with text-embedding-3-large)
• Pinecone account (for vector storage)

Installation

Via npx (recommended)

No installation needed. Configure directly in your MCP client:

{
  "mcpServers": {
    "cf-codebase": {
      "command": "npx",
      "args": ["-y", "coldfusion-mcp"],
      "env": {
        "OPENAI_API_KEY": "sk-...",
        "PINECONE_API_KEY": "pcsk_...",
        "PINECONE_INDEX_NAME": "cf-codebase",
        "REDIS_URL": "redis://localhost:6379",
        "CF_CODEBASE_PATH": "/path/to/your/coldfusion/codebase"
      }
    }
  }
}

Global install

npm install -g coldfusion-mcp

Then configure your MCP client to use the coldfusion-mcp command.

Configuration

All configuration is via environment variables passed through your MCP client's env block:

| Variable | Required | Description | |----------|----------|-------------| | CF_CODEBASE_PATH | Yes | Path to the ColdFusion codebase root | | OPENAI_API_KEY | Yes | OpenAI API key for embeddings | | PINECONE_API_KEY | Yes | Pinecone API key for vector storage | | PINECONE_INDEX_NAME | Yes | Pinecone index name (e.g., cf-codebase) | | REDIS_URL | No | Redis connection URL (default: redis://localhost:6379) | | CF_MANIFEST_PATH | No | Path to codebase-manifest.json (auto-detected if not set) | | CF_INCLUDE_GLOB | No | File patterns to index (default: **/*.{cfm,cfc}) | | CF_EXCLUDE_GLOB | No | File patterns to skip (default: **/node_modules/**,**/.git/**) |

Tools

| Tool | Description | |------|-------------| | search_codebase | Semantic search across the CF codebase using vector embeddings | | trace_dependencies | Trace structural dependencies (callers/callees) of a component | | get_component_api | Get the full public API of a ColdFusion component | | find_sql_usage | Find all components that query a given database table | | explain_component | Get an LLM-generated explanation of what a component does | | find_similar_code | Find existing code similar to a description to prevent duplicates | | get_index_status | Get health and freshness status of the search index | | refresh_index | Trigger re-indexing of the ColdFusion codebase |

Initial Setup

1. Start Redis:

   docker compose up -d redis

2. Initialize the Pinecone index:

   npx coldfusion-mcp  # or: node dist/server.js

   Then use the refresh_index tool from your MCP client to build the initial index.

Development

git clone https://github.com/swa34/coldFusion-MCP.git
cd coldFusion-MCP
npm install
cp .env.example .env  # fill in your API keys
docker compose up -d redis
npm run dev

How It Works

• Tree-sitter parses ColdFusion files into ASTs using CFML grammar (WASM)
• OpenAI generates embeddings with text-embedding-3-large (3072 dimensions)
• Pinecone stores and queries vectors for semantic search
• Redis caches the dependency graph and component metadata
• The MCP server exposes all analysis capabilities as callable tools

License

MIT