Overview
What is Contextual MCP Server?
A Model Context Protocol (MCP) server that provides Retrieval-Augmented Generation (RAG) capabilities using Contextual AI. It bridges AI interfaces like Cursor IDE and Claude Desktop to a dedicated Contextual AI agent, enabling query processing, intelligent retrieval from a knowledge base, and context-aware responses with citations.
How to use Contextual MCP Server?
Install with Python 3.10 or higher, clone the repo, create a virtual environment, and run pip install -e .. Configure by setting your Contextual AI API key and agent ID in a .env file (or directly), then modify the MCP configuration file (.cursor/mcp.json or Claude Desktop config) with the path to uv and the server script. Invoke the query tool from any MCP client.
Key features of Contextual MCP Server
- Query processing via a dedicated Contextual AI agent
- Intelligent retrieval from a comprehensive knowledge base
- Context-aware responses grounded in source documentation
- Responses include citations and attributions
- Maintains conversation context for follow-up questions
- Flexible tooling; new tools can be added with
@mcp.tool()
Use cases of Contextual MCP Server
- Domain-specific research, e.g., financial data on large US firms
- Technical documentation queries, e.g., semiconductor datasheets
- Code retrieval and explanation within an IDE like Cursor
- General knowledge-base Q&A with source attribution
FAQ from Contextual MCP Server
What prerequisites are needed?
Python 3.10 or higher, Cursor IDE and/or Claude Desktop, a Contextual AI API key, and an MCP-compatible environment.
How do I configure the API key and agent ID?
Set the API_KEY and AGENT_ID environment variables in a .env file or directly in your shell before starting the server.
Where should I place the MCP configuration file?
For Cursor, project-specific: .cursor/mcp.json; global: ~/.cursor/mcp.json. For Claude Desktop, use the same format in its configuration directory.
Can I extend the server with my own tools?
Yes. Create functions decorated with @mcp.tool(), define parameters with Python type hints, and provide a clear docstring describing the tool's purpose.
What are the current limitations?
The server runs locally and may not work in remote development environments. Tool responses are subject to Contextual AI API limits and quotas. Only stdio transport mode is currently supported.
