Best MCP servers for Cursor in 2026, with mcp.json configs

The best MCP servers for Cursor cover what your codebase can't tell the agent: Context7 for current library docs, GitHub for issues and PRs, Playwright or Chrome DevTools to see the page you're building, a database server, your ticket tracker and error monitor, and a source of outside data. Each takes a few lines in .cursor/mcp.json. We checked every server below against its own repo or docs on October 17, 2026, and each has a config you can paste.

Fewer servers work better than more, so each entry says when to skip it. One of them, ChatterAPI, is ours. It sits in the section where it fits.

How Cursor MCP config works

Cursor's MCP docs (checked October 17, 2026) describe two config files:

• .cursor/mcp.json in your project, "for project-specific tools"
• ~/.cursor/mcp.json in your home directory, "for tools available everywhere"

The two files use the same format. A local server runs a command:

{
  "mcpServers": {
    "server-name": {
      "command": "npx",
      "args": ["-y", "some-mcp-server"],
      "env": { "API_KEY": "value" }
    }
  }
}

A remote server is a URL, plus headers if it needs a key:

{
  "mcpServers": {
    "server-name": {
      "url": "https://example.com/mcp",
      "headers": { "Authorization": "Bearer ${env:EXAMPLE_KEY}" }
    }
  }
}

Cursor supports stdio, SSE and Streamable HTTP transports, and OAuth for servers that require it.

Keep keys out of the file

Cursor resolves variables in command, args, env, url and headers: ${env:NAME} from your environment, plus ${userHome}, ${workspaceFolder}, ${workspaceFolderBasename} and ${pathSeparator}. If you put ${env:...} in the headers, you can commit .cursor/mcp.json and your team gets the same servers while each person keeps their own key in their shell.

Installing from the marketplace

Editing JSON is optional. The docs say you can "Click 'Add to Cursor' on a marketplace entry to install it and authenticate with OAuth." Official plugins live in the Cursor Marketplace, and community servers are listed on cursor.directory. Several of the servers below also publish their own Add to Cursor button.

The 10 servers at a glance

Star counts come from the GitHub API, checked on October 17, 2026.

Docs and code

1. Context7

Context7 (MIT) pulls current documentation and examples for the libraries you use into the agent's context, so it stops suggesting APIs that were renamed two versions ago. The fastest setup is npx ctx7 setup --cursor. To configure it by hand, use the URL from the README and a Bearer key from the Context7 dashboard:

{
  "mcpServers": {
    "context7": {
      "url": "https://mcp.context7.com/mcp",
      "headers": { "Authorization": "Bearer ${env:CONTEXT7_API_KEY}" }
    }
  }
}

2. GitHub

GitHub's Cursor install guide recommends the hosted server and needs Cursor v0.48.0 or newer. It also notes that in Cursor "the GitHub server currently requires a Personal Access Token":

{
  "mcpServers": {
    "github": {
      "url": "https://api.githubcopilot.com/mcp/",
      "headers": { "Authorization": "Bearer ${env:GITHUB_PAT}" }
    }
  }
}

You can skip it if you only work in one repo and the agent's terminal with gh covers you.

The browser

3. Playwright MCP

Playwright MCP (Apache-2.0) lets the agent open your app, click through it and read the page. The standard config from its README:

{
  "mcpServers": {
    "playwright": {
      "command": "npx",
      "args": ["@playwright/mcp@latest"]
    }
  }
}

4. Chrome DevTools MCP

Chrome DevTools MCP (Apache-2.0) gives the agent DevTools itself: console errors, network requests and performance traces. For Cursor, its guide says to go to Cursor Settings, MCP, New MCP Server and paste the standard config:

{
  "mcpServers": {
    "chrome-devtools": {
      "command": "npx",
      "args": ["-y", "chrome-devtools-mcp@latest"]
    }
  }
}

Start with one of 3 and 4. Pick Playwright for testing flows, or DevTools when you need to find out why a page is slow or broken.

Data and project tools

5. Supabase

Supabase's server is hosted. The README says your client "will automatically prompt you to log in to Supabase during setup." Options go in the URL, so you can scope it to one project and turn on read-only mode:

{
  "mcpServers": {
    "supabase": {
      "url": "https://mcp.supabase.com/mcp?project_ref=<project-ref>&read_only=true"
    }
  }
}

Start read-only, and let the agent write to your database only once you've decided it should.

6. Linear

Linear's MCP docs say you can install it from a one-click link or by searching for Linear on Cursor's MCP tools page. To add it by hand, it's https://mcp.linear.app/mcp with OAuth, or https://mcp.linear.app/mcp/readonly for read-only tools:

{
  "mcpServers": {
    "linear": { "url": "https://mcp.linear.app/mcp" }
  }
}

7. Sentry

Sentry's hosted server lives at https://mcp.sentry.dev/mcp (repo getsentry/toolkit) and signs you in with OAuth. Its README also shows passing a token in a Sentry-Bearer header for clients that support custom headers:

{
  "mcpServers": {
    "sentry": { "url": "https://mcp.sentry.dev/mcp" }
  }
}

The web and social data

8. Exa

Exa's server (MIT) adds web search and page fetching, hosted at https://mcp.exa.ai/mcp. Its README says "OAuth is preferred: most clients prompt you to sign in to Exa," and that you can limit the tools in the URL with ?tools=.

{
  "mcpServers": {
    "exa": { "url": "https://mcp.exa.ai/mcp" }
  }
}

Skip it if Cursor's built-in web search already answers your questions.

9. Firecrawl

Firecrawl's server (MIT) scrapes and crawls pages into clean content. Its Cursor instructions run it locally with a key:

{
  "mcpServers": {
    "firecrawl-mcp": {
      "command": "npx",
      "args": ["-y", "firecrawl-mcp"],
      "env": { "FIRECRAWL_API_KEY": "${env:FIRECRAWL_API_KEY}" }
    }
  }
}

Its README makes a point that applies to every server on this list: "Never put an API key in the server URL. Never put an API key in an agent chat."

10. ChatterAPI (Reddit and X)

Web search won't give you a Reddit thread's comments or an X account's recent posts in structured form. ChatterAPI is our hosted MCP server for that. It has 36 tools: 34 for Reddit and X (search posts and comments, read threads, subreddits, users, X profiles, timelines, replies and followers) and two for your account. Reddit is read-only, and X actions work only from accounts you connect. It's independent and not affiliated with Reddit or X.

{
  "mcpServers": {
    "chatterapi": {
      "url": "https://api.chatterapi.com/mcp",
      "headers": { "Authorization": "Bearer ${env:CHATTERAPI_KEY}" }
    }
  }
}

Open Cursor Settings, then MCP, and the server shows its tool count once it connects. We measured one thing on October 17, 2026: our tools/list response describes those 36 tools in 106,378 bytes of JSON. That makes it a server to enable in the projects that need it, not globally. A real reddit_search call through it returned 20 posts for 2 credits. Here's one of them, trimmed:

{
  "id": "1slr3qj",
  "title": "MCP vs API?",
  "author": "SelectionNo3613",
  "subreddit": "ClaudeAI",
  "score": 12,
  "numComments": 26,
  "createdAt": "2026-04-15T00:52:13.000Z"
}

Skip it if your project never touches Reddit or X. For open-source single-platform options, see the best MCP servers for Reddit and X.

A starter .cursor/mcp.json

Three servers cover most of what a typical web app repo needs. Commit this file, and have each developer set the variables in their own shell:

{
  "mcpServers": {
    "context7": {
      "url": "https://mcp.context7.com/mcp",
      "headers": { "Authorization": "Bearer ${env:CONTEXT7_API_KEY}" }
    },
    "github": {
      "url": "https://api.githubcopilot.com/mcp/",
      "headers": { "Authorization": "Bearer ${env:GITHUB_PAT}" }
    },
    "playwright": {
      "command": "npx",
      "args": ["@playwright/mcp@latest"]
    }
  }
}

Add the database, tracker and data servers only in the projects that use them.

Approvals and safety

Cursor's docs say it "asks for approval before using MCP tools by default," and MCP tools follow the same run modes as terminal commands. In Auto-review mode, allowlisted tools run straight away and everything else goes through a classifier. The docs also warn: "Remember that MCP servers can access external services and execute code on your behalf." Their advice is to check the source, review permissions, limit API keys and audit the code of critical integrations.

What we do in practice:

• Pick the official server for a service over a community wrapper of it.
• Run in read-only mode where a server offers one (Supabase and Linear do) until you trust the setup.
• Leave approvals on for tools that write, meaning anything that posts, sends or deletes.
• Read tools bring outside text into the session. A web page, a Reddit comment or an issue body can hold instructions aimed at the agent, so read what it proposes before you approve it.

Do you need MCP in Cursor at all?

Not always. Cursor's agent can run terminal commands, so for a lot of single-project work a CLI, or a curl call with the docs in your rules file, is enough. MCP earns its place when the same tools have to work in other clients too, when no good CLI exists, or when you want the agent to discover and pick tools on its own. MCP vs API compares the two approaches with a worked example. If you also use Claude Code, the best MCP servers for Claude Code has the claude mcp add commands for most of these.

Add Reddit and X to Cursor

Sign up with Google for 100 free credits (you don't need a card), paste the ChatterAPI entry above into .cursor/mcp.json, and ask the agent what Reddit says about a library you're evaluating. Setup for other clients is on the MCP server page. Plans start at $19 a month for 25,000 credits.