project-context-map-mcp
Generates structured project memory and exposes MCP tools for AI coding agents to query topology, read relevant files, and log changes in a repository.
README
Project Context Map MCP
<div align="center">
Give your coding agent a map before it starts wandering
Turn a repository into structured project memory for MCP clients like Claude Code, so they can query context, inspect only relevant files, and understand topology before touching code.
</div>
Why This Exists
AI coding tools often begin the same way in a medium or large repo:
- they read too many files
- they spend tokens on unrelated areas
- they rebuild context from scratch every time
- they miss the real dependency chain behind a feature or bug
project-context-map-mcp fixes that by generating a lightweight project memory and exposing it through MCP tools.
Instead of:
scan everything -> guess -> open random files
you get:
index first -> query topology -> read sparse files -> edit with context
What This Project Does
This package gives you:
- an MCP server for project-aware queries
- a CLI to generate and refresh repository memory
- a readable
.project.mdwith topology highlights and change log - a machine-readable
.project-map.json - a human-readable
.project-map.md - import/dependency relationships between files
- sparse file access limited to indexed paths
- MCP-assisted writes with automatic logging
- Claude Code setup helpers and skill generation
At A Glance
| Part | Purpose |
|---|---|
.project.md |
Main readable project memory for humans and agents |
.project-map.json |
Structured index used by tools |
.project-map.md |
Lightweight markdown summary |
read_project_context |
Returns .project.md |
query_topology |
Finds relevant files from natural-language queries |
read_files_sparse |
Reads only indexed files |
write_and_log |
Writes changes and appends to .project.md Change Log |
Installation
Option 1: Global install
Best if you want to use it across many repositories.
npm install -g project-context-map-mcp
Option 2: Local install
Best if you want it only inside one project.
npm install project-context-map-mcp
Requirements
- Node.js
20or newer - an MCP-compatible client such as Claude Code
Quick Start
1. Generate the project memory
Run this from the root of the repo you want to index:
project-context-map-mcp refresh --project-root .
This creates:
.project.md.project-map.json.project-map.md
2. Configure Claude Code for that repo
project-context-map-mcp configure-claude --project-root .
This creates or updates:
.mcp.json.claude/skills/project-context-map/SKILL.md
3. Start the MCP server
If your client starts the server from .mcp.json, you usually do not need to run it manually.
If you want to run it yourself:
project-context-map-mcp serve --project-root .
4. Use it from your MCP client
Typical flow:
- read the project context
- query topology with a natural-language task
- read only the matched files
- make changes
- refresh the map
Prompt examples:
- See PROMPTS.md for tested starter prompts for Claude Code, Cursor, and VS Code Copilot Chat.
- For best results, use prompts that name the feature area, module, or source files you want the agent to inspect.
MCP Config
If you want to wire the server manually, add this to your MCP config:
{
"mcpServers": {
"project-context-map": {
"command": "project-context-map-mcp",
"args": ["serve", "--project-root", "."],
"env": {}
}
}
}
Generate it automatically
project-context-map-mcp print-mcp-config --project-root .
Write it into the repo
project-context-map-mcp configure-claude --project-root .
CLI Commands
project-context-map-mcp serve --project-root .
project-context-map-mcp refresh --project-root .
project-context-map-mcp install-hooks --project-root .
project-context-map-mcp configure-claude --project-root .
project-context-map-mcp print-mcp-config --project-root .
project-context-map-mcp help
<details> <summary><strong>Command details</strong></summary>
| Command | What it does |
|---|---|
serve |
Starts the MCP server over stdio |
refresh |
Regenerates .project.md, .project-map.json, and .project-map.md |
install-hooks |
Installs git hooks to keep the map refreshed |
configure-claude |
Creates repo-local Claude Code skill and .mcp.json |
print-mcp-config |
Prints the MCP config JSON without writing files |
help |
Shows command help |
</details>
MCP Tools
These are the main tools exposed by the server.
Core workflow tools
| Tool | Purpose |
|---|---|
read_project_context |
Returns .project.md content |
query_topology |
Takes a natural-language query and returns matched file paths from the query index |
read_files_sparse |
Reads file content only for indexed paths |
write_and_log |
Writes file changes and appends an entry to .project.md Change Log |
Additional tools
| Tool | Purpose |
|---|---|
get_project_summary |
Returns summary, stack, modules, hotspots |
refresh_project_map |
Regenerates the project memory artifacts |
get_module_context |
Returns a focused view of one module |
find_relevant_files |
Legacy ranking-oriented file query |
get_recent_git_changes |
Summarizes recent git activity |
explain_file_role |
Explains a file’s role and recent changes |
<details> <summary><strong>What each core tool returns</strong></summary>
read_project_context
Returns:
.project.mdpath- full markdown content
query_topology
Returns:
- matched file paths
- confidence score
- match reasons
- per-file dependencies
- per-file
usedByrelationships
read_files_sparse
Returns:
- file content for indexed files
- safe errors for files outside the repo or outside the known topology
write_and_log
Returns:
- path written
- bytes written
- appended change log entry
</details>
Recommended Workflow
flowchart LR
A[Refresh project map] --> B[Read .project.md]
B --> C[Query topology]
C --> D[Read sparse files]
D --> E[Implement change]
E --> F[Write and log]
F --> G[Refresh project map]
Example task flow
User asks:
Fix the login bug where session expires after refresh
Recommended agent behavior:
- call
read_project_context - call
query_topologywith the task - call
read_files_sparsewith the matched file paths - inspect only those files first
- edit code
- call
write_and_logif using MCP-managed writes - call
refresh_project_map
Topology & Visual Graph
The project map is not only a file list. It also extracts internal relationships such as:
- JavaScript and TypeScript imports
- CommonJS
require(...)usage - basic Python import detection
- reverse usage information with
used_by
That data is written into:
.project-map.json.project.md
.project.md includes a Mermaid graph so markdown viewers that support Mermaid can render a visual dependency map.
What It Generates
.project.md
The main working document. It includes:
- project summary
- key modules
- query index preview
- topology highlights
- Mermaid dependency graph
- running change log
.project-map.json
The structured source of truth. It includes fields such as:
project_summarytech_stackmodulesfilestopologyentrypointsdependenciesrecent_changeshotspots
.project-map.md
A shorter human-readable summary.
Claude Code Setup
After running:
project-context-map-mcp configure-claude --project-root .
your repo will contain:
.mcp.json
.claude/skills/project-context-map/SKILL.md
The generated skill tells Claude Code to:
- consult project memory first
- avoid broad repo scans
- query relevant files before opening source
- refresh the project map after edits
Git Hooks
If you want the map to stay fresh automatically:
project-context-map-mcp install-hooks --project-root .
This installs:
post-commitpost-mergepost-push
Each hook refreshes the project memory after repository activity.
Interactive Setup Guide
<details open> <summary><strong>I just installed it. What should I do next?</strong></summary>
- Go to your repo root.
- Run
project-context-map-mcp refresh --project-root . - Run
project-context-map-mcp configure-claude --project-root . - Open the repo in Claude Code.
- Approve the MCP server from
.mcp.json. - Ask the agent to use
read_project_contextandquery_topologyfirst.
</details>
<details> <summary><strong>I want to use my own MCP config instead of the generated one</strong></summary>
Use:
project-context-map-mcp print-mcp-config --project-root .
Then copy the JSON into your own MCP config file and adjust:
- server name
- command path
- working project root
</details>
<details> <summary><strong>I only want the package inside one repo</strong></summary>
Install locally:
npm install project-context-map-mcp
Then run through npx:
npx project-context-map-mcp refresh --project-root .
npx project-context-map-mcp configure-claude --project-root .
</details>
<details> <summary><strong>How do I know it is working?</strong></summary>
You should see:
.project.md.project-map.json.project-map.md.mcp.jsonafter Claude setup
You can also run:
project-context-map-mcp help
project-context-map-mcp refresh --project-root .
</details>
Example Prompts For Your MCP Client
Tested prompt templates now live in PROMPTS.md.
Quick examples:
Use the project context map first, then find the files involved in session refresh logic.Use query_topology to locate the files related to npm publishing, package configuration, and release safety checks.Use get_module_context for src and explain how the CLI and MCP server fit together.Find the files that control authentication and prefer source files over docs.
Best Fit Repositories
- Node.js applications
- React apps
- Next.js projects
- mixed frontend/backend repos
- Python services
- large internal tools where AI agents are used often
Local Development
npm install
npm run check
node src/cli.js refresh --project-root .
node src/cli.js serve --project-root .
Current Limitations
- topology detection is strongest for JS, TS, and simple Python import patterns
- visual graph output is Mermaid markdown, not a full interactive web app
- natural-language ranking quality depends on file names, summaries, tags, and recent changes
License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.