Test Coverage MCP
Make your agents coverage-aware as they code for you
“Hey, I’m a coding agent. I just created flashy nifty feature… but oops, I downgraded the coverage 🤓. How could I know that?”
“Hey, I’m a testing agent. I was tasked to cover some code with testing, but how can I find which areas are not covered?😳”
Give your coding and testing agent eyes: MCP server that provides instant, reliable, token-efficient test coverage data for any programming language (LCOV based)
__
🚀 Just launched (November 2025) ! I spend great time these days on polishing this library. If you find this valuable, a ⭐ star helps signal to other developers that this project is worth their attention
The Problem
When AI coding agents work on your code without proper coverage tooling, they face three critical issues:
Coverage Blindness - They can't see if their changes improved or regressed test coverage
Token Waste - They burn thousands of tokens trying to parse massive LCOV files (some exceed 10 MB)
Unreliable Scripts - They improvise custom parsing scripts that often fail or produce incorrect results
The Solution
This MCP server solves all three problems by providing:
Coverage Awareness - Agents can check coverage anytime with a simple tool call
Token Efficiency - Get coverage summaries in <100 tokens instead of thousands
Accuracy - Production-grade LCOV parsing that handles all format variations
Baseline Tracking - Measure coverage progress within a session without keeping state in memory
Test Coverage: This project maintains 95% test coverage and we're targeting 100% soon.
Two Main Workflows
1. Query Coverage Summary
Ask for overall project coverage or coverage for specific files:
2. Baseline Tracking for Session Progress
Establish a baseline at session start, then measure your progress:
Why baseline tracking? Without it, agents would need to keep initial coverage in their stateful memory throughout the session, consuming valuable context window space.
Installation
Configuration
Add this MCP server to your AI coding tool's configuration:
Claude Desktop (Claude Code)
macOS: Edit ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: Edit %APPDATA%\Claude\claude_desktop_config.json
After updating, restart Claude Desktop.
Cursor IDE
Create or edit .cursor/mcp.json in your project root:
GitHub Copilot (VS Code)
Create or edit .vscode/mcp.json in your workspace:
Requires VS Code 1.99+ or Visual Studio 17.14+. Enterprise users need "MCP servers in Copilot" policy enabled.
Windsurf (Codeium IDE)
macOS: Edit ~/.codeium/windsurf/mcp_config.json
Windows: Edit %APPDATA%\Codeium\Windsurf\mcp_config.json
Linux: Edit ~/.codeium/windsurf/mcp_config.json
Or use the GUI: Settings → Advanced Settings → Cascade → Add Server
Available Tools
coverage_summary
Get overall project coverage from an LCOV file.
Input:
Output:
Example:
coverage_file_summary
Get coverage for a specific file.
Input:
Output:
Example:
start_recording
Record current coverage as a baseline for later comparison.
Input:
Output:
Example:
get_diff_since_start
Compare current coverage against the recorded baseline.
Input:
Output:
Example:
Usage Examples
Example 1: Check Coverage Before Starting Work
Example 2: Track Coverage Impact During Development
Example 3: Verify Specific File Coverage
How It Works
This MCP server:
Parses LCOV files using a production-grade parser that handles all LCOV format variations
Calculates coverage percentages for overall project or individual files
Stores baselines in a temporary directory for session-based tracking
Returns compact JSON responses that consume minimal tokens
LCOV Format Support
This server supports all standard LCOV file formats, including:
Files with summary sections (
SF:,end_of_record)Files with line-by-line data only (
DA:entries)Files with branch coverage data (
BRDA:,BRF:,BRH:)Mixed formats within the same file
Troubleshooting
"LCOV file not found"
Ensure you've run your test suite with coverage enabled first
Check that the path to your LCOV file is correct (relative paths are resolved from current working directory)
Default path is
./coverage/lcov.info
"No coverage data found for file"
Verify the file path matches exactly as it appears in the LCOV file
Some test frameworks use absolute paths, others use relative paths
"No baseline recording found"
You must call
start_recordingbefore callingget_diff_since_startBaselines are stored in temporary storage and cleared when the system restarts
Development
Contributing
Contributions are welcome! Please feel free to submit a Pull Request.
License
MIT © Yoni Goldberg
Links
Improvement ideas
coverage_file_summary returns nested properties also declared as flat
Start recording overrides other sessions files
Improve record naming - setSessionBaseline, getDiffSinceBaseline
This server cannot be installed