Why these MCP servers
The developer MCP stack is the most mature part of the ecosystem. GitHub MCP and Git MCP expose the version control layer. Filesystem MCP gives the agent local access. JetBrains MCP and Playwright MCP add IDE and browser tooling. Sentry and Datadog MCPs bring production observability into the editor. Together, they turn a coding agent into a teammate that can read code, open pull requests, debug production errors, and ship.
Recommended MCP servers
A curated shortlist of the most useful MCP servers for this workflow.
GitHub MCP Server
Repository intelligence inside agent sessions
The official GitHub MCP server exposes repositories, issues, pull requests, code search, and Actions context to coding agents. It is published and maintained by GitHub and is the most widely adopted MCP server in production.
Filesystem MCP
Controlled local file access for AI agents
A reference MCP server that lets assistants read, write, and manage approved local directories with explicit boundaries and safe path handling. It is the canonical example for any agent that needs structured access to a developer machine.
Git MCP
Read, search, and manipulate local repositories
Expose common Git operations to MCP clients: status, log, diff, blame, branch management, and commit history search. A perfect companion to coding agents that work on local clones.
Playwright MCP
Reliable browser automation for AI agents
A browser automation MCP server built on Playwright. Agents can navigate pages, click elements, fill forms, capture screenshots, and inspect the DOM with deterministic waits and accessibility-aware locators.
JetBrains MCP
IDE-aware tools for coding agents
Bridge JetBrains IDEs to MCP clients with project awareness, refactoring actions, and code intelligence. The companion plugin works with IntelliJ, PyCharm, WebStorm, and GoLand.
Docker MCP
Container and compose for agents
Manage Docker containers, images, and Compose stacks from MCP clients. A practical primitive for agents that provision services and run side-effect tools.
Sentry MCP
Errors, releases, and stack traces for agents
Surface Sentry issues, releases, and stack traces inside agent sessions so debugging tools can propose fixes against real error context. Read-only by default with optional write scopes.
Datadog MCP
Metrics, logs, and traces for agents
Surface Datadog metrics, logs, and APM traces inside MCP clients. A foundation for SRE and observability agents that reason about production systems.
How to set this up
- 1
Install Claude Desktop or Cursor
Both ship with first-class MCP support. Claude Desktop uses claude_desktop_config.json; Cursor has a UI for adding servers under Settings → Features → Model Context Protocol.
- 2
Add GitHub MCP
The most important server for any coding agent. It exposes repository files, issues, pull requests, and Actions context. Use a fine-grained personal access token with the repos, issues, and pull_requests scopes.
- 3
Add Filesystem MCP
Gives the agent local file access inside an approved directory. Always scope the path argument to the project root, not your home directory.
- 4
Add Playwright MCP
For any task that requires verifying UI in a real browser. Playwright MCP works with Chromium, Firefox, and WebKit.
- 5
Add observability MCPs (optional)
Sentry MCP and Datadog MCP surface production errors and metrics inside the agent. Useful for debugging and post-mortem work.
Frequently asked questions
What is the minimum MCP setup for a coding agent?
A useful starting point is GitHub MCP + Filesystem MCP + Git MCP. Add Playwright MCP for browser verification and Sentry MCP for production error context.
Should I use npx or Docker for MCP servers?
npx is faster to set up and works well for development. Docker is more reproducible and gives you better isolation. We recommend npx for local development and Docker for production or shared environments.