Best MCP servers for Cursor, grouped by the job they do

The best MCP servers for Cursor give its agent context it cannot get from the open files. Cursor already indexes your workspace and searches it. An MCP (Model Context Protocol) server is worth adding when it brings in something new: how your code connects, your pull requests, current library docs, a browser or production errors. Graphify, a code knowledge graph for AI coding agents, covers the first of those jobs.

Pick by job, not by popularity. Each server's tools load into the agent's context, so a short list beats a long one.

Which MCP servers are best for Cursor?Link to Which MCP servers are best for Cursor?#

For most developers, a good Cursor set is a code graph for structure, GitHub for pull requests, Context7 for library docs, and Playwright for UI checks. Add Sentry when you debug production and a read-only Postgres server when you work in SQL. Keep the active set to three to five.

| Job | Example server | Runs | Add it when | Skip it when | | --- | --- | --- | --- | --- | | Code structure | Graphify | Hosted (OAuth) or local (stdio) | You ask "what calls this?" or "what breaks if I change it?" | The repo is small enough to read | | Pull requests and issues | GitHub MCP | Hosted (OAuth) | Your team works in GitHub | You rarely leave the editor for GitHub | | Library docs | Context7 | Hosted or local | Your libraries ship breaking changes | Your dependencies are stable | | Browser | Playwright MCP | Local | You want the agent to test its UI | You build no frontend | | Errors | Sentry MCP | Hosted (OAuth) | You triage production errors | You do not use Sentry |

Dupple and Parallel publish similar shortlists, and both warn against running too many servers at once.

Why add a code graph if Cursor already indexes my repo?Link to Why add a code graph if Cursor already indexes my repo?#

Cursor's index finds code that looks like your question. A code graph finds code that is connected to it. Those are different questions. "Where is retry logic?" suits search. "What calls charge_card, and what calls those?" needs the call graph, because callers rarely share words with the function they call.

Graphify is a code knowledge graph for AI coding agents. It parses code with tree-sitter and records calls, imports and references with file paths. Cursor's agent asks one structural question and gets a small, connected answer. Our post on code knowledge graphs vs RAG explains the difference in more depth.

How do I add an MCP server to Cursor?Link to How do I add an MCP server to Cursor?#

Add it to ~/.cursor/mcp.json for every project, or .cursor/mcp.json for one repository. A remote server needs only its URL, and Cursor opens a browser sign-in the first time it connects. You can also install many servers in one click from Cursor's MCP settings. Hosted Graphify looks like this:

{
  "mcpServers": {
    "graphify": {
      "url": "https://api.graphify.com/mcp"
    }
  }
}

Open Cursor's MCP settings. The server shows "Needs login" until you sign in through the browser with OAuth. Then start a new agent chat and paste this test prompt:

Use Graphify to list the repositories you can see. If the tool fails, report the error instead of guessing.

Connect a repository at app.graphify.com before you test. The Cursor integration docs cover project scope and a default repository.

Can Cursor use a local code graph instead?Link to Can Cursor use a local code graph instead?#

Yes. The open-source Graphify CLI ships a local MCP server, graphify-mcp, that runs over stdio. It needs no account and keeps code on your machine. Open-source Graphify is licensed under Apache 2.0. Build a graph with the CLI, then point Cursor at it with an absolute path:

{
  "mcpServers": {
    "graphify": {
      "command": "graphify-mcp",
      "args": ["/absolute/path/to/repo/graphify-out/graph.json"]
    }
  }
}

Install it first with uv tool install "graphifyy[mcp]". The local server has ten tools, such as query_graph, get_neighbors and shortest_path. See the MCP server quickstart.

What should I check before installing any MCP server?Link to What should I check before installing any MCP server?#

Check who maintains it, what it can write and what credentials it needs. Prefer servers from the vendor whose service they wrap. Give tokens the narrowest scope that works, and use read-only modes for databases. Cursor's own MCP docs give the same advice: verify the source and limit API keys.

Then watch your tool count. If the agent starts calling the wrong tool, turn servers off until it stops.

Try itLink to Try it#

Sign in at app.graphify.com, connect a repository and add the mcp.json entry above. To stay local, run uv tool install graphifyy and graphify install --platform cursor. Our guide on giving Cursor a code knowledge graph covers what to ask first.

Related reading: what a code knowledge graph is, giving Cursor a code knowledge graph and codebase context tools compared.

FAQLink to FAQ#

How many MCP servers should Cursor run?Link to How many MCP servers should Cursor run?#

Three to five active servers is a good range. More tools make the agent slower to choose and more likely to pick the wrong one.

Where does Cursor store MCP settings?Link to Where does Cursor store MCP settings?#

In ~/.cursor/mcp.json for all projects, or .cursor/mcp.json inside a repository for that project only.

Does Graphify work in Cursor without an API key?Link to Does Graphify work in Cursor without an API key?#

Yes. The hosted server uses OAuth sign-in in the browser. The local server reads a graph file on your disk.

Is a code graph the same as Cursor's codebase index?Link to Is a code graph the same as Cursor's codebase index?#

No. The index finds similar text. A code graph follows calls, imports and references between files, so it finds connected code.

Can I commit Cursor's MCP config for my team?Link to Can I commit Cursor's MCP config for my team?#

Yes. Commit .cursor/mcp.json with the hosted URL. Each teammate signs in with their own account.