Cursor AI · Curated ranking
Best MCP Servers for Cursor in 2026
By Ethan Halfhide · Updated October 20, 2026
BestMCPTools
Best MCP Servers for Cursor in 2026
bestmcptools.org
Last verified 2026-10-20
The best MCP servers for Cursor in October 2026 are Context7 for current library docs, Playwright MCP and Chrome DevTools MCP for driving a real browser, Supabase MCP for your database, and Sentry MCP for production errors, with Exa, Firecrawl and Fetch covering web research. All eight run in Cursor today, either from a one-click install or a few lines in mcp.json. This guide covers what is specific to Cursor: where the config lives, project versus global scope, OAuth and tool approval. Most picks sit in our developer tools category or a close neighbour, and every fact comes from the vendor's own docs or repo.
TL;DR, Quick Verdict
Start with Context7 so Cursor's agent stops guessing at library APIs, then add one browser server (Playwright MCP for scripted flows, Chrome DevTools MCP for debugging and performance traces). If your app runs on Supabase, add its hosted server scoped to one project in read-only mode, and add Sentry MCP scoped to one project if you ship to production. Put team-wide servers in .cursor/mcp.json and commit it; put anything that needs your personal token in ~/.cursor/mcp.json. Every pick has a free way in.
Quick Picks
Context7 MCP
Pulls version-specific docs into Cursor so generated code matches the API you actually use.
Playwright MCP
Drives pages through accessibility snapshots, so no vision model is needed.
Chrome DevTools MCP
Gives the agent network requests, console logs and performance traces from a live Chrome.
Supabase MCP
Hosted server with OAuth login, project scoping and a read-only mode.
Sentry MCP
Remote server built for coding agents, scoped to an org or project.
Fetch MCP
Open source, no account, turns any public page into markdown.
How we chose
We compared official information only: each server's own repository README, the vendor's setup docs, and public pricing pages, all read on October 20, 2026. We did not test these servers. A server made the list if its maker actively maintains it, it documents a config Cursor accepts, and it solves an editor problem: docs, browser checks, database queries or production errors. We left out servers whose own maintainers point users elsewhere. The reference GitHub server that many older lists recommend now lives in the archived modelcontextprotocol servers-archived repository, which states it receives no security updates; GitHub maintains its own server at github/github-mcp-server instead.
How Cursor loads MCP servers
Cursor reads MCP servers from two files. A project file at .cursor/mcp.json applies only to that repository, and Cursor's help docs suggest committing it so teammates get the same tools. A global file at ~/.cursor/mcp.json applies to every project you open. Cursor merges both, and if the same server name appears in each, the project entry wins. You can also skip the file entirely: the Customize page in the sidebar lists servers from the Cursor Marketplace with an Add to Cursor button, and community servers are listed on cursor.directory. Several picks here (Context7, Playwright, Chrome DevTools, Exa) publish their own Add to Cursor deeplinks in their READMEs or client setup guides. After editing mcp.json by hand, Cursor's help page says to save and restart Cursor.
Project vs global config
The project file is the right home for servers the whole team should share and that carry no personal secret, such as Context7, Playwright MCP, Chrome DevTools MCP or a Supabase entry that relies on OAuth. Keep personal tokens out of it. Cursor's mcp.json supports interpolation in the command, args, env, url and headers fields, so a committed entry can read ${env:SENTRY_ACCESS_TOKEN} or ${workspaceFolder} instead of a literal value. Local stdio servers can also load an envFile such as ${workspaceFolder}/.env, but Cursor's docs note that envFile does not work for remote HTTP or SSE servers. Use the global file for tools you want everywhere, like Fetch MCP or a search server tied to your own API key.
Which one fits your project
For a frontend or full stack web app, the strongest Cursor pairing is Context7 plus one browser server. Choose Playwright MCP server when you want the agent to click through flows and turn them into tests, and Chrome DevTools MCP server when you want it to read the console, inspect network calls and record performance traces. If your backend is Supabase, Supabase MCP server lets the agent list tables, run SQL and fetch logs. Sentry MCP server brings production issues into the chat. For research that goes beyond docs, pick Exa MCP server for search, Firecrawl MCP server for scraping and crawling, or Fetch MCP server for single pages with no account. More browser options are in the browser automation category.
Local vs remote setup
Cursor supports three transports: stdio, where Cursor starts a local command and manages it for one user, and SSE or Streamable HTTP, where you point Cursor at a URL and it can authenticate with OAuth. In mcp.json a local server uses command and args; a remote one uses url and optional headers. Playwright MCP, Chrome DevTools MCP and Fetch MCP run locally. Supabase, Sentry and Exa recommend their hosted endpoints, and Context7 and Firecrawl offer both. Remote servers need no local install, but requests pass through the vendor. Local servers keep traffic on your machine, which Cursor's own FAQ recommends for sensitive data. Cursor Cloud Agents take MCP servers from the Cloud Agents dashboard rather than your local file.
Costs and accounts
As of October 2026, Context7's Free plan includes 1,000 API calls per month and Pro is $10 per seat per month. Exa's free Starter tier gives $10 of credits every month plus a $10 onboarding bonus. Firecrawl's Free plan includes 1,000 credits per month, its hosted keyless endpoint allows scrape, search and parse without a key, and the Hobby plan is $16 per month billed yearly. Supabase MCP is free to connect, and Supabase's Free plan is $0 with two active projects while Pro starts at $25 per month. Sentry MCP needs a Sentry account; we did not check Sentry plan prices for this page. Playwright MCP, Chrome DevTools MCP and Fetch MCP are free open source tools with no account.
Security and permissions in Cursor
By default Cursor asks for approval before each MCP tool call. Cursor's help docs say that from Cursor 3.6 the default Auto-review mode runs allowlisted MCP tools immediately and sends everything else through a safety classifier, and a permissions.json file can preconfigure which tools skip approval. Supabase's own docs recommend scoping the server to one project with project_ref, using read_only=true, limiting feature groups, and never giving the server to end users because it acts with your developer permissions. Sentry recommends scoping its URL to a single project. On Enterprise plans, admins can allowlist approved MCP commands, URLs and individual tools from the Cursor dashboard.
Keeping the tool list manageable
Every enabled server adds tools the agent has to consider. Firecrawl's full profile registers 27 tools, while Exa enables two by default. Chrome DevTools MCP has a --slim mode and Supabase's features parameter trims tool groups. In Cursor you can toggle a server off from Customize without deleting its config, or turn single tools off from the tools list at the top of the chat panel. When something fails, open the Output panel and choose MCP Logs.
What we could not verify
We looked for a documented cap on how many MCP tools Cursor's agent can use at once, but the Cursor MCP docs and help pages we read on October 20, 2026 do not state one, so we do not quote a number. We also did not confirm Sentry plan pricing or how each server performs on large codebases, since we compared documentation rather than running the servers. If you are choosing between Cursor and a terminal agent, our guide to MCP servers to add to Claude Code covers the same ground for that client.
Quick Comparison
| Server | Best for | Transport | Auth | Maintainer | Price (as of October 2026) |
|---|---|---|---|---|---|
| Context7 MCP | Current library docs | Remote HTTP or local stdio | API key header (recommended) or OAuth via ctx7 setup | Upstash | Free 1,000 calls/month; Pro $10 per seat/month |
| Playwright MCP | Scripted browser flows | Local stdio | None | Microsoft | Free, open source |
| Chrome DevTools MCP | Debugging and performance traces | Local stdio | None | Chrome DevTools team (Google) | Free, open source |
| Supabase MCP | Supabase database and logs | Remote HTTP | OAuth login | Supabase | Server free; Supabase Free $0, Pro from $25/month |
| Sentry MCP | Production errors and traces | Remote HTTP or local stdio | OAuth, or Sentry user auth token for stdio | Sentry | Server free; Sentry account required |
| Exa MCP | AI web search | Remote Streamable HTTP | Anonymous with rate limits, OAuth or API key | Exa Labs | Free $10 credits/month; search from $4 per 1k requests |
| Firecrawl MCP | Scraping and crawling | Remote HTTP or local stdio | Keyless tier, OAuth or API key | Firecrawl | Free 1,000 credits/month; Hobby $16/month billed yearly |
| Fetch MCP | Single pages to markdown | Local stdio | None | Model Context Protocol project | Free, open source |
The Tools, Ranked
Context7 MCP
Context7, from Upstash, fetches up-to-date, version-specific documentation and code examples into the agent's context through two tools, resolve-library-id and query-docs. Its docs recommend ~/.cursor/mcp.json, or .cursor/mcp.json for one project, and offer Add to Cursor buttons for remote and local versions. Context7 also suggests a rule under Cursor Settings, Rules, telling the agent to use it for library documentation automatically.
Pros
- ✓Remote endpoint means nothing to install
- ✓Free API key raises rate limits
- ✓Works for any library in the Context7 index
Cons
- ✗Free plan stops after 1,000 API calls per month, with 20 bonus calls a day
- ✗Library docs are community contributed and Context7 does not guarantee their accuracy
- ✗API backend and crawler are not open source
Quick Install
{
"mcpServers": {
"context7": {
"url": "https://mcp.context7.com/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_KEY"
}
}
}
}Example
Ask Cursor to wire up a Next.js route handler and let it pull the current Next.js docs before writing code.
Microsoft's Playwright MCP controls a browser through Playwright's structured accessibility snapshots rather than screenshots, so it does not need a vision model. The README includes an Add to Cursor button and a manual path: add a command type server with npx @playwright/mcp@latest. It needs Node.js 18 or newer and runs headed unless you pass --headless.
Pros
- ✓Deterministic, accessibility tree based actions
- ✓One-click Cursor install
- ✓Can record actions you perform in the browser and return them as Playwright code
Cons
- ✗Accessibility snapshots can be large in the context window
- ✗Concurrent clients need --isolated or separate profiles
Quick Install
{
"mcpServers": {
"playwright": {
"command": "npx",
"args": ["@playwright/mcp@latest"]
}
}
}Example
Have Cursor sign up a test user on localhost, then turn the steps into a Playwright test file.
Chrome DevTools MCP, published by the Chrome DevTools team, gives the agent a live Chrome with access to network requests, screenshots, console messages with source-mapped stack traces, and performance traces. It starts the browser only when a tool needs it, and its client guide has an Add to Cursor button. It officially supports Google Chrome and Chrome for Testing. Usage statistics collection is on by default; pass --no-usage-statistics to opt out.
Pros
- ✓Real DevTools data, including performance insights
- ✓--slim mode for basic tasks
- ✓One-click Cursor install
Cons
- ✗Usage statistics on by default
- ✗Officially supports Chrome only
- ✗Exposes browser content to the MCP client
Quick Install
{
"mcpServers": {
"chrome-devtools": {
"command": "npx",
"args": ["-y", "chrome-devtools-mcp@latest"]
}
}
}Example
Ask Cursor why the checkout page is slow and let it record a trace and point at the blocking request.
Supabase MCP
Supabase's hosted server at mcp.supabase.com connects Cursor to your projects so the agent can list tables, run SQL, apply migrations, read logs and fetch advisors. Supabase's setup docs tell Cursor users to add the URL to .cursor/mcp.json, or to ~/.cursor/mcp.json for every project, and Cursor prompts you to sign in with OAuth. URL parameters let you scope it with project_ref, force read_only=true, and limit features. Supabase recommends those limits, plus branching, before you point it at real data.
Pros
- ✓No local install and OAuth login
- ✓Project scoping and read-only mode
- ✓Destructive SQL confirmation when the client supports elicitation
Cons
- ✗Acts with your developer permissions
- ✗Self-hosted and CLI setups get fewer tools
- ✗Confirmation prompts are a guardrail, not a guarantee
Quick Install
{
"mcpServers": {
"supabase": {
"url": "https://mcp.supabase.com/mcp?project_ref=YOUR_PROJECT_REF&read_only=true"
}
}
}Example
Ask Cursor which tables lack row level security and draft a migration for review.
Sentry's remote MCP server is built for human-in-the-loop coding agents and names Cursor as a target client. All hosted connections use OAuth, and Sentry's setup page gives a Cursor snippet plus an install button. You can scope the URL to an organization or a single project, which Sentry recommends, and discovery tools are hidden when scoped. A stdio mode with a user auth token, still a work in progress, covers self-hosted Sentry.
Pros
- ✓Hosted with OAuth
- ✓Org and project scoping
- ✓Works with self-hosted Sentry over stdio
Cons
- ✗AI search tools need an LLM provider
- ✗Stdio mode still a work in progress
- ✗Requires a Sentry account
Quick Install
{
"mcpServers": {
"sentry": {
"url": "https://mcp.sentry.dev/mcp/YOUR_ORG/YOUR_PROJECT"
}
}
}Example
Paste a Sentry issue link and ask Cursor to find the stack frame in your code and propose a fix.
Exa MCP
Exa's hosted server at mcp.exa.ai exposes web_search_exa and web_fetch_exa by default, with advanced search and Exa Agent available through a tools URL parameter. The README has an Add to Cursor button. The hosted server works anonymously with rate limits, and OAuth or an API key raises limits and unlocks Exa Agent.
Pros
- ✓Works without an account
- ✓Small default tool set
- ✓One-click Cursor install
Cons
- ✗Anonymous use is rate limited
- ✗Paid usage after free credits
Quick Install
{
"mcpServers": {
"exa": {
"url": "https://mcp.exa.ai/mcp"
}
}
}Example
Ask Cursor to find recent migration notes for a dependency before it upgrades your lockfile.
Firecrawl MCP
Firecrawl's server scrapes, searches, maps and crawls websites. The hosted endpoint allows scrape, search and parse without a key, and OAuth or an API key unlocks the full tool set; Firecrawl says never to put a key in the URL. The README also documents a local Cursor setup with npx -y firecrawl-mcp and FIRECRAWL_API_KEY in env. With default settings the full profile registers 27 tools, so consider the keyless or search-only endpoint if you want fewer.
Pros
- ✓Keyless hosted tier
- ✓Local or remote
- ✓Search-only endpoint with a fixed tool set
Cons
- ✗Large tool count in the full profile
- ✗Crawl, map and agent tools need a key
Quick Install
{
"mcpServers": {
"firecrawl-mcp": {
"command": "npx",
"args": ["-y", "firecrawl-mcp"],
"env": {
"FIRECRAWL_API_KEY": "YOUR-API-KEY"
}
}
}
}Example
Have Cursor crawl a vendor's API reference and save it as markdown in your docs folder.
Fetch is a reference server from the official Model Context Protocol servers repository. Its one tool, fetch, turns a URL into markdown and can read long pages in chunks with start_index. It is a Python server, and the README recommends running it with uvx mcp-server-fetch. It obeys robots.txt for model-initiated requests and by default refuses private, loopback and cloud metadata addresses, though the README still cautions about local network access.
Pros
- ✓No account or key
- ✓Single small tool
- ✓Blocks private addresses by default
Cons
- ✗No search, only known URLs
- ✗Needs uv or Python installed
Quick Install
{
"mcpServers": {
"fetch": {
"command": "uvx",
"args": ["mcp-server-fetch"]
}
}
}Example
Ask Cursor to read a changelog page and summarise breaking changes before you bump a version.
The Verdict
For most Cursor users, Context7 plus one browser server is the right first install, and both have one-click Add to Cursor buttons. Add Supabase MCP or Sentry MCP only when your stack uses them, and scope each to a single project.
Decide on placement before you install: shared, secret-free servers go in the committed .cursor/mcp.json, personal tokens go in ~/.cursor/mcp.json or come from ${env:} variables. For the wider field, see our MCP server directory and the search category.
Frequently Asked Questions
Can Cursor use MCP servers?
Yes. Cursor supports MCP servers over stdio, SSE and Streamable HTTP. You can install them from the Customize page with Add to Cursor, or define them in .cursor/mcp.json for one project or ~/.cursor/mcp.json for all projects.
What are the best MCP servers for Cursor for free?
Playwright MCP, Chrome DevTools MCP and Fetch MCP are free open source tools with no account. Context7, Exa and Firecrawl have free tiers with monthly limits, and Supabase MCP is free to connect to a free Supabase project.
Where is the Cursor MCP config file?
Project servers go in .cursor/mcp.json inside the repository and global servers go in ~/.cursor/mcp.json in your home folder. Cursor merges both, and a project entry wins if the names match.
Which MCP server is best for developers using Cursor?
Context7 is the most broadly useful because it gives the agent current library docs. After that, pick a browser server such as Playwright MCP or Chrome DevTools MCP, then add servers for the services your app uses.
How do I fix an MCP server that will not connect in Cursor?
Open the Output panel, choose MCP Logs and look for startup or authentication errors. Then toggle the server off and on in Customize or restart Cursor.
Are MCP servers still useful in Cursor?
Yes, for tasks that need live data such as current docs, your database, production errors or a running browser. Some vendors, including Microsoft, also offer CLI and skills options that use fewer tokens.
Sources
- Cursor docs: Model Context Protocol
- Cursor help: MCP integrations
- Context7 repository
- Context7 client setup
- Context7 plans and pricing
- Playwright MCP repository
- Chrome DevTools MCP repository
- Chrome DevTools MCP client configurations
- Supabase MCP setup docs
- Supabase MCP repository
- Supabase pricing
- Sentry MCP server
- Sentry MCP repository
- Exa MCP repository
- Exa pricing
- Firecrawl MCP repository
- Firecrawl pricing
- Fetch MCP server
- Archived MCP reference servers
- GitHub MCP Server repository
About the author
Ethan Halfhide
Ethan Halfhide is an AI technologist who works with AI agents and Model Context Protocol integrations. He curates BestMCPTools' server directory, checking each server's documentation, supported clients and setup steps.