Skip to content

Best MCP Servers for Cursor in 2026: The Complete Setup Guide

The 10 best MCP servers for Cursor — from code search to tool recommendations. Setup instructions included.

March 24, 2026
10 min read

Cursor added MCP support in late 2025 and has been improving it since. In January 2026, a major update introduced dynamic context management across MCP servers — reducing token usage by 47% when using multiple servers simultaneously.

But most guides dump a list of servers without explaining how they fit into a real Cursor workflow. This guide covers the 12 MCP servers that actually matter for Cursor users, with the exact configs and workflows to set them up.

How MCP Works in Cursor

Cursor reads MCP server definitions from two places:

Global (all projects): ~/.cursor/mcp.json

Per-project: .cursor/mcp.json in the project root

Both use the same format. Servers defined at the project level override global ones with the same name.

{
  "mcpServers": {
    "server-name": {
      "command": "npx",
      "args": ["-y", "package-name"],
      "env": {
        "API_KEY": "your_key"
      }
    }
  }
}

After saving, go to Cursor Settings > Tools & MCP to verify your servers loaded. Each server shows a green/red status indicator. If a server fails, check the error message — it is usually a missing env var or Node.js not in PATH.

Transport options: Cursor supports stdio (local subprocess), SSE (Server-Sent Events), and Streamable HTTP. For most MCP servers, stdio is the right choice — it runs locally, is fast, and needs zero infrastructure. Use Streamable HTTP only when connecting to hosted remote servers.

The 12 Best MCP Servers for Cursor

Tier 1: Install These First

1. Context7 MCP

Problem it solves: Cursor's AI cites outdated library docs from its training data. You are using Next.js 15, but it generates code for Next.js 13 patterns.

Context7 fetches the current documentation for whatever library you are working with — at call time, not training time. It resolves version-specific APIs so the AI generates code against today's docs, not last year's.

When you will use it: Every time you write code using a framework or library. Especially valuable with fast-moving projects (Next.js, SvelteKit, Tailwind, Prisma).

"context7": {
  "command": "npx",
  "args": ["-y", "@upstash/context7-mcp"]
}

No API key needed. Works immediately.

2. GitHub MCP Server

Problem it solves: You tab-switch to GitHub 20 times a day — to check issues, review PR comments, search code across repos, or create a branch.

The GitHub MCP server lets Cursor do all of this inline. "Create a PR for the changes I just made" or "Search all repos for how we handle auth" — Claude calls the GitHub API and returns results in the conversation.

When you will use it: PRs, issue tracking, cross-repo code search, branch management.

"github": {
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-github"],
  "env": {
    "GITHUB_TOKEN": "ghp_your_token"
  }
}

Create a fine-grained personal access token with only the repos and permissions you need. Do not use a classic token with full access.

3. Toolradar MCP

Problem it solves: "What's the best headless CMS?" or "Compare Prisma vs Drizzle" — Cursor's AI guesses from training data. The answers are often outdated or wrong, especially for pricing.

Toolradar MCP gives Cursor access to 8,400+ software tools with verified pricing (checked weekly), G2/Capterra ratings, editorial scores, and AI-identified alternatives. The AI searches, compares, and recommends based on real data — not hallucinated data.

When you will use it: Choosing a library, service, or tool for your project. Evaluating pricing. Comparing alternatives before committing to a dependency.

"toolradar": {
  "command": "npx",
  "args": ["-y", "toolradar-mcp"],
  "env": {
    "TOOLRADAR_API_KEY": "tr_live_your_key"
  }
}

Free key (100 calls/day): toolradar.com/dashboard/api-keys

Tier 2: Add Based on Your Stack

4. PostgreSQL MCP Server

For: Backend developers with Postgres databases.

"Show me the schema of the orders table" or "How many users signed up this week?" — the AI generates SQL, runs it, and returns results. Use a read-only database user for safety.

"postgres": {
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-postgres"],
  "env": {
    "DATABASE_URL": "postgresql://readonly_user:pass@host:5432/db"
  }
}

5. Playwright MCP

For: Frontend developers who need to test or debug UI.

"Test the login flow" generates and runs a Playwright test. "Take a screenshot of the homepage after login" does what you'd expect. Useful for quick visual verification without setting up a full test suite.

"playwright": {
  "command": "npx",
  "args": ["-y", "@playwright/mcp"]
}

No API key. Requires Playwright browsers installed (npx playwright install).

6. Docker MCP Server

For: Anyone running services in Docker locally.

"Spin up a Postgres container on port 5433" or "Show me the logs from the api container" — manages your local dev environment from the AI chat. Also generates Dockerfiles.

"docker": {
  "command": "npx",
  "args": ["-y", "mcp-server-docker"]
}

7. Brave Search MCP

For: When you need current information the LLM does not have.

"How does Next.js 15 handle server actions?" gets live web results, not training-data answers. Independent from Google, privacy-respecting.

"brave-search": {
  "command": "npx",
  "args": ["-y", "@brave/brave-search-mcp-server"],
  "env": {
    "BRAVE_API_KEY": "your_key"
  }
}

Free tier with $5/month credits at brave.com/search/api.

8. Sentry MCP

For: Teams using Sentry for error monitoring.

"What's the most frequent error this week?" or paste an error ID and get the full stack trace, breadcrumbs, and affected users — all in the conversation. Debugging goes from "reproduce the issue" to "fix the issue" directly.

"sentry": {
  "command": "npx",
  "args": ["-y", "@sentry/mcp-server"],
  "env": {
    "SENTRY_AUTH_TOKEN": "your_token"
  }
}

Tier 3: Specialized

9. Figma MCP

For: Frontend developers implementing designs.

Exposes your Figma selection's live structure — hierarchy, auto-layout, spacing tokens, component refs. Cursor generates code that matches the real design, not a screenshot interpretation.

10. Vercel MCP

For: Teams deploying on Vercel.

"Why did my last deploy fail?" gets the actual build log. Manage env vars, check deployment status, configure domains — without the dashboard.

11. Linear MCP

For: Teams using Linear for project management.

Create issues, manage sprints, search across projects — all from Cursor. "Create a bug ticket for the redirect issue, assign to me, add to current sprint."

12. Firecrawl MCP

For: When you need to scrape or read web content.

Converts any webpage to clean, LLM-ready markdown. Useful for reading documentation, extracting data, or research that Brave Search results alone cannot satisfy.

Copy this into ~/.cursor/mcp.json:

{
  "mcpServers": {
    "context7": {
      "command": "npx",
      "args": ["-y", "@upstash/context7-mcp"]
    },
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_TOKEN": "ghp_your_token"
      }
    },
    "toolradar": {
      "command": "npx",
      "args": ["-y", "toolradar-mcp"],
      "env": {
        "TOOLRADAR_API_KEY": "tr_live_your_key"
      }
    },
    "brave-search": {
      "command": "npx",
      "args": ["-y", "@brave/brave-search-mcp-server"],
      "env": {
        "BRAVE_API_KEY": "your_key"
      }
    }
  }
}

These four cover: current docs (Context7), code management (GitHub), tool intelligence (Toolradar), and live web (Brave Search). Add database, testing, and deployment servers as your workflow demands.

Cursor-Specific Tips

Dynamic context (January 2026 update). Cursor now intelligently manages context across all connected MCP servers, reducing token usage by 47%. This means having 5-6 servers connected is practical — the overhead is much lower than before.

Enable only what you need. In Settings > Tools & MCP, you can toggle individual servers on/off without removing the config. Disable servers you are not actively using to reduce noise.

Project-level configs. Put .cursor/mcp.json in your project root for project-specific servers. A web project might have Playwright and Vercel; a data project might have PostgreSQL and E2B. Commit the file (without secrets) so your team shares the same MCP setup.

Secrets management. Never put API keys directly in .cursor/mcp.json if the file is committed to git. Use environment variables from your shell:

{
  "env": {
    "GITHUB_TOKEN": "${GITHUB_TOKEN}"
  }
}

Then export the variable in your .zshrc or .bashrc.

Troubleshooting. If a server shows red status: (1) check that npx works in your terminal, (2) verify the env vars are set, (3) try running the server manually: npx -y package-name — if it errors, that is your issue.

What is Coming

Cursor's MCP support is evolving fast. Based on the Cursor Community Forum and recent updates:

  • MCP tool auto-enable — ability to pre-configure which tools are enabled in mcp.json instead of manually toggling after each restart
  • MCP Marketplace — Cursor launched a marketplace at cursor.com/marketplace for one-click MCP installation (live since February 2026)
  • Improved error handling — better feedback when MCP servers fail silently

The ecosystem is moving fast. With 97 million monthly MCP SDK downloads and 146 organizations in the Agentic AI Foundation, expect the tooling to improve rapidly.

Get your Toolradar API key: toolradar.com/dashboard/api-keys →

Full list of servers: 25 best MCP servers for 2026 →

New to MCP? What is an MCP server? →

mcpcursormcp-serversdeveloper-toolsbest-ofsetup
Share this article