MCP Server List: Best MCP Servers for AI Coding Tools

What Is An MCP Server?

An MCP server exposes a controlled capability to an AI coding tool. Instead of pasting context manually, the tool can inspect a repo, query a database, read a ticket, or drive a browser through a defined interface.

Best Categories

Useful MCP servers tend to map to repeated developer context: code, files, databases, design, browser state, and planning tools.

• GitHub: Issues, pull requests, code search, release notes
• Filesystem: Reading and editing project files with scoped access
• Postgres: Schema inspection, SQL exploration, query debugging
• Supabase: Supabase projects, auth tables, edge-function workflows
• Figma: Design context, screenshots, tokens, and asset handoff
• Browser: Opening local apps, clicking flows, visual checks
• Playwright: End-to-end test runs and browser screenshots
• Linear: Issue triage, project status, implementation planning
• Notion: Specs, docs, decision records, research synthesis
• Search: Docs lookup, package details, troubleshooting references

Starter mcp.json

Keep the first config small and review every command or token boundary before sharing it with a team.

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "."]
    }
  }
}

How To Evaluate MCP Servers

A good MCP server should save repeated context gathering, reduce manual copying, and expose only the access the AI coding tool truly needs. Avoid adding servers simply because they are available.

• Prefer read-only access for databases and docs unless write actions are required.
• Keep credentials out of committed config files.
• Document who owns each server and how to rotate tokens.
• Test one server at a time before sharing the config with a team.

Recommended Starter Stack

Most AI coding setups should start with a small MCP stack. Filesystem provides local repo context, GitHub connects issues and pull requests, and Playwright or Browser adds UI verification when the project has a frontend.

• Minimal backend stack: Filesystem + GitHub.
• Frontend stack: Filesystem + GitHub + Playwright or Browser.
• Product/design stack: Filesystem + GitHub + Figma.
• Data-heavy stack: Filesystem + GitHub + read-only Postgres.

Security And Permission Review

MCP servers can expose powerful context and actions, so treat them like developer tools with real access. Document scopes, owners, tokens, and whether the server can read, write, or execute anything outside the project.

• Use least privilege for tokens and service accounts.
• Prefer read-only access until a write action is clearly needed.
• Keep secrets in environment variables, not committed mcp.json files.
• Review server access again when a repo moves from prototype to production.

MCP Config Review Template

Before copying a generated MCP config into a team repository, review it with the same care as any other developer environment file.

Review checklist:
[ ] Does each server have a clear owner?
[ ] Are tokens stored outside the repo?
[ ] Is database access read-only where possible?
[ ] Is the command available in GUI-launched tools?
[ ] Is the fallback workflow documented?

FAQ