Skip to main content

​
MCP Server Overview

OpenGround provides a Model Context Protocol (MCP) server that enables AI agents to search your local documentation database in real-time.

​
What is MCP?

The Model Context Protocol is an open protocol that standardizes how AI applications provide context to Large Language Models (LLMs). It enables secure, controlled access to data sources through a client-server architecture.

​
How OpenGround Uses MCP

OpenGround implements an MCP server (openground-mcp) that exposes three powerful tools to AI agents:

search_documents_tool

Search documentation using semantic search and BM25 hybrid retrieval

list_libraries_tool

View all available libraries and versions in your local database

get_full_content_tool

Retrieve complete page content from search results

​
Server Architecture

The MCP server runs as a subprocess managed by your AI agent. Communication happens over stdio (standard input/output), which provides:
  • Security: No network exposure, runs locally only
  • Performance: Direct process communication with zero network latency
  • Simplicity: No ports, no authentication, no TLS configuration

​
Benefits

​
1. Real-Time Documentation Access

Instead of relying on the AI’s training data (which may be outdated), agents can query the latest official documentation from your local database.

​
2. Isolated Context

Documentation searches run in a separate context, preventing your main conversation from being cluttered with search results.

​
3. Offline Operation

Once documentation is indexed, all searches work completely offline—no API calls, no rate limits.

​
4. Version-Specific Results

Search documentation for specific library versions, ensuring compatibility with your project dependencies.

​
5. Multi-Library Support

Index and search documentation from multiple frameworks simultaneously: 
# Add multiple libraries
openground add react --source https://github.com/facebook/react.git --docs-path docs -y
openground add fastapi --source https://github.com/tiangolo/fastapi.git --docs-path docs -y
openground add django --source https://github.com/django/django.git --docs-path docs -y

​
MCP Tool Details

​
search_documents_tool

Performs hybrid search (semantic + BM25) against your local documentation database. Parameters:
  • query (string): The search query
  • library_name (string): Which library to search (e.g., “react”, “fastapi”)
  • version (string): Library version (e.g., “latest”, “v18.2.0”)
Returns: Markdown-formatted results with relevant chunks, scores, and source URLs

​
list_libraries_tool

Lists all documentation libraries and their versions available in your local database. Parameters: None Returns: Dictionary mapping library names to lists of available versions 
{
  "react": ["latest"],
  "fastapi": ["latest", "v0.109.0"],
  "django": ["latest", "v5.0"]
}

​
get_full_content_tool

Retrieves the complete content of a documentation page by URL. Parameters:
  • url (string): The page URL from search results
  • version (string): Library version
Returns: Full page content in markdown format

​
Agent Workflow

When you ask an AI agent about a library, the typical workflow is:
1

Check Available Libraries

Agent calls list_libraries_tool to see what documentation is available
2

Search Documentation

Agent calls search_documents_tool with your query, library name, and version
3

Present Results

Agent shows you relevant snippets with source URLs
4

Fetch Full Content (Optional)

If needed, agent calls get_full_content_tool to get complete page content

​
Server Configuration

The MCP server uses your OpenGround configuration: 
# View current configuration
openground config show

# Configure search settings
openground config set query.top_k 10

# Configure embedding model
openground config set embeddings.model_name "sentence-transformers/all-MiniLM-L6-v2"

​
Performance Optimizations

The openground-mcp server includes several optimizations:
  1. Background initialization: Embedding models are loaded in a background thread
  2. Cache warming: Metadata is pre-cached on startup
  3. Minimal logging: Stdout pollution is suppressed for clean stdio communication

​
Next Steps

Install for Claude Code

Set up OpenGround with Claude’s AI agent

Install for Cursor

Configure OpenGround in Cursor’s mcp.json

Install for OpenCode

Add OpenGround to OpenCode’s config

Custom Agents

Manual configuration for other MCP clients