ArchestackGitHub ↗ (opens in a new tab)

MCP servers for developers

MCP servers that give coding agents tools: docs lookup, browsers, GitHub, databases and code navigation.

codegraph leads with 80, ahead of chrome-devtools-mcp (64) and codebase-memory-mcp (62). 6 projects ranked by score.

The ranking

MCP servers for developers: full ranking
RankProjectAdoptionFreshnessMaintenanceEasy to runAgent-readyScore
1codegraphLocal code knowledge graph served to coding agents over MCP73.7k stars, MIT, last commit Oct 20269210091508580 out of 100
2chrome-devtools-mcpMCP server that lets coding agents drive and debug a live Chrome53.2k stars, Apache-2.0, last commit Oct 20266910080333064 out of 100
3codebase-memory-mcpMCP server that indexes codebases into a queryable knowledge graph46.3k stars, MIT, last commit Oct 2026471008250062 out of 100
#4context7Fetches current, version-specific library docs into coding agent prompts62.9k stars, MIT, last commit Oct 202680100990057 out of 100
#5github-mcp-serverMCP server exposing GitHub repos, issues, PRs and Actions to AI tools33.5k stars, MIT, last commit Oct 2026321008533054 out of 100
#6cli-anythingGenerates command-line wrappers so AI agents can drive desktop and backend software51.8k stars, Apache-2.0, last commit Sep 202659100430043 out of 100

Momentum, Verified build, Docs and Privacy are not measured yet; their weight goes to the signals shown. A dash means the signal is not scored for that kind of project. Hover a number for its rating in words.

Reviews

180 out of 100

codegraph

Local code knowledge graph served to coding agents over MCP

73.7k stars, MIT, last commit Oct 2026

CodeGraph indexes a project into a local graph of symbols, call edges and dependencies under `.codegraph/`, then exposes it to coding agents through an MCP server. Agents can fetch relevant source, call paths and change impact in one call instead of grepping and reading files. `codegraph install` wires it into Claude Code, Cursor, Codex CLI, opencode, Gemini CLI, Kiro, GitHub Copilot and others, and auto-sync keeps the index current.

Strengths

  • Runs locally; no Node.js needed, ships as a bundled installer or npm package
  • One `codegraph install` configures MCP for nine named agents and IDEs
  • Covers 30+ languages, including TypeScript, Python, Go, Rust, Java, C#, Swift
  • Watches files and updates the graph automatically; no manual re-index

Weaknesses

  • README admits about 80% more retrieval context stays resident in long sessions
  • Published benchmarks are the author's, on 7 repos with one question each
  • No Docker or compose setup; install is via curl/PowerShell script or npm
  • Hosted CodeGraph platform is a waitlist product, separate from this repo
  • no GPU
Documentation ↗ (opens in a new tab)Website ↗ (opens in a new tab)Repository on GitHub ↗ (opens in a new tab)
264 out of 100

chrome-devtools-mcp

MCP server that lets coding agents drive and debug a live Chrome

53.2k stars, Apache-2.0, last commit Oct 2026

chrome-devtools-mcp is an MCP server that gives coding agents such as Claude, Cursor and Copilot control of a live Chrome browser. It records performance traces through Chrome DevTools, inspects network requests and console messages with source-mapped stack traces, and automates page actions via puppeteer. A CLI is included for use without MCP, and a --slim mode exposes a smaller tool set.

Strengths

  • Records DevTools performance traces and extracts insights from them
  • Puppeteer-based automation waits for action results automatically
  • Slim mode limits the exposed tools to basic browser tasks
  • Also ships a CLI for use without an MCP client

Weaknesses

  • Usage statistics are sent to Google by default; opt out with --no-usage-statistics
  • Performance tools may send trace URLs to the Google CrUX API
  • Officially supports only Chrome and Chrome for Testing
  • Exposes all browser content to the connected MCP client
  • no GPU
  • Needs Node.js LTS, npm, Google Chrome (current stable)
Documentation ↗ (opens in a new tab)Repository on GitHub ↗ (opens in a new tab)
362 out of 100

codebase-memory-mcp

MCP server that indexes codebases into a queryable knowledge graph

46.3k stars, MIT, last commit Oct 2026

A native C executable that parses repositories with tree-sitter (162 languages) into a persistent graph of functions, classes, call chains, HTTP routes and cross-service links, and exposes it through 17 MCP tools. It adds semantic search with bundled embeddings, Cypher-like queries, dead code detection, git diff impact mapping and a built-in 3D graph UI. Processing is local, and `install` configures detected coding-agent clients.

Strengths

  • Runs locally with no API key, hosted service or language runtime
  • Bundled embeddings enable semantic search without Ollama or Docker
  • Covers 162 languages; Hybrid LSP type resolution for 10 of them
  • Releases are VirusTotal-scanned and SLSA 3 attested

Weaknesses

  • Updates require re-running the install script; `update` only prints the command
  • Defender may flag release binaries as a false positive, per the README
  • All active processes must share the exact same build and cache root
  • Watcher covers only git projects opened in MCP sessions, by default
  • no GPU
  • Models: nomic-embed-code (bundled, int8, 768d)
  • port 9749
Documentation ↗ (opens in a new tab)Repository on GitHub ↗ (opens in a new tab)
#457 out of 100

context7

Fetches current, version-specific library docs into coding agent prompts

62.9k stars, MIT, last commit Oct 2026

Context7 looks up documentation and code examples for a named library and version, then returns them to the coding agent as context. It runs as an MCP server with two tools (resolve-library-id, query-docs) or as a ctx7 CLI plus skill. Setup is one command, npx ctx7 setup, which handles OAuth and API key generation. The hosted service at mcp.context7.com does the retrieval.

Strengths

  • Two modes: MCP server, or ctx7 CLI with an installed skill
  • Library IDs like /vercel/next.js skip matching; version can be named in the prompt
  • npx ctx7 setup configures Cursor, Claude Code or opencode
  • Also ships a TypeScript SDK and Vercel AI SDK tools

Weaknesses

  • API backend, parsing and crawling engines are private; repo has only the MCP server
  • Relies on the hosted service at mcp.context7.com; local run is a separate guide
  • Free API key recommended for higher rate limits; exact limits unknown
  • Docs are community-contributed; accuracy and security are not guaranteed
  • no GPU
  • Needs Node.js 18+, Context7 hosted API
Documentation ↗ (opens in a new tab)Website ↗ (opens in a new tab)Repository on GitHub ↗ (opens in a new tab)
#554 out of 100

github-mcp-server

MCP server exposing GitHub repos, issues, PRs and Actions to AI tools

33.5k stars, MIT, last commit Oct 2026

Official GitHub MCP server, written in Go, that lets MCP hosts read code, manage issues and pull requests, inspect Actions runs and review security alerts. It runs either as a GitHub-hosted remote endpoint (api.githubcopilot.com/mcp/) or locally over stdio via the ghcr.io Docker image or a binary built with go build. Authentication is OAuth or a personal access token.

Strengths

  • Hosted remote endpoint needs no local install; local Docker image and Go build also available
  • Supports OAuth login or PAT; PAT takes precedence when set
  • Works with GitHub Enterprise Cloud (ghe.com) and Enterprise Server via GITHUB_HOST
  • Install guides for VS Code, Claude, Cursor, Windsurf, Zed, Codex and others

Weaknesses

  • Enterprise Server is not supported by the remote server; local server only
  • Remote OAuth needs each host to configure a GitHub App or OAuth App
  • Local Docker OAuth needs a fixed callback port (8085) published to loopback
  • Only talks to GitHub; no other forges or providers
  • no GPU
  • Docker
  • Needs GitHub API, Docker (optional)
Documentation ↗ (opens in a new tab)Repository on GitHub ↗ (opens in a new tab)
#643 out of 100

cli-anything

Generates command-line wrappers so AI agents can drive desktop and backend software

51.8k stars, Apache-2.0, last commit Sep 2026

CLI-Anything runs a 7-phase generator inside a coding agent such as Claude Code, Cursor or Codex. It produces a Python (Click) CLI harness with JSON output and a SKILL.md for a target application or codebase. A companion package, cli-anything-hub, installs community-built CLIs from a registry, covering GIMP, Blender, LibreOffice, QGIS, n8n and others.

Strengths

  • Ready-made registry of community CLIs installable with `cli-hub install <name>`
  • Generated CLIs emit structured JSON and ship a SKILL.md for agent discovery
  • Works with several agent hosts: Claude Code, Codex, Cursor, OpenClaw, Pi, OpenCode
  • Apache-2.0 license; README reports 2,461 passing tests

Weaknesses

  • Many CLIs only wrap an upstream app, which must be installed separately
  • Generating a new CLI requires a supported AI coding agent, so output quality varies
  • Harness quality is per-CLI and community-contributed; test depth differs between harnesses
  • Windows use with Claude Code needs Git for Windows or WSL for bash and cygpath
  • no GPU
  • Needs Python 3.10+, A supported AI coding agent (for generating new CLIs), Upstream apps the chosen CLI wraps (e.g. GIMP, Blender, LibreOffice)
Documentation ↗ (opens in a new tab)Website ↗ (opens in a new tab)Repository on GitHub ↗ (opens in a new tab)

Written from each project's README and checked facts. Spot something wrong? Report it on GitHub (opens in a new tab).