Best MCP Servers for Claude Code in 2026: 10 for Real Dev Work

The best MCP servers for Claude Code in 2026 are GitHub for code, Context7 for current library docs, Playwright for the browser, Sentry for errors, Vercel and Cloudflare for hosting, Supabase and Neon for databases, Figma for designs and Stripe for payments.

Claude Code is Anthropic's coding agent for the terminal and IDE. You add a server with claude mcp add, and Claude Code can then call that server's tools while it works. Claude Code's MCP docs recommend remote HTTP servers and mark the older SSE transport as deprecated.

Every server on this list is official, and we checked each against its vendor's docs on 28 October 2026. eSEOspace did not test these servers.

How we chose

We included a server only if it passed these tests:

1. Official. Published by the product's own company or its official GitHub organization.
2. Documented for Claude Code. The vendor gives a Claude Code command or a standard HTTP setup that Claude Code supports.
3. Live and documented. We opened its docs on 28 October 2026.
4. Clear safety story. The docs explain sign-in and, where relevant, read-only modes or scoping.
5. Clear cost. Open source, or plan rules stated in the docs.

We focused on what a developer touches daily: code, docs, tests, errors, hosting and data. Our developer servers list covers similar tools across all clients; this list is about Claude Code setup and safety.

Comparison table

1. GitHub MCP Server: repos, issues and pull requests

What it is. The GitHub MCP Server is GitHub's official server, MIT licensed. GitHub's Claude Code guide adds the remote server at https://api.githubcopilot.com/mcp with a personal access token, using claude mcp add-json on Claude Code 2.1.1 or later.

The --toolsets flag picks which tool groups load. The README says "write tools are skipped if --read-only is set."

Best for: Any team whose code lives on GitHub.

Pricing: GitHub's docs say it "is available to all GitHub users regardless of plan type." Tools for paid features need that feature's subscription. See our GitHub connect guide.

Pros

• Free on every plan.
• Read-only mode overrides everything else.
• Toolsets keep the tool list short.

Cons

• The remote setup in the Claude guide uses a token, not OAuth.
• Run the setup command in your normal terminal, not inside Claude Code.

Checked against: GitHub MCP Claude install guide on 28 October 2026.

2. Context7: current docs for any library

What it is. Context7, from Upstash, "pulls up-to-date, version-specific documentation and code examples straight from the source." It has two MCP tools: resolve-library-id and query-docs.

The remote URL is https://mcp.context7.com/mcp. npx ctx7 setup --claude sets it up for Claude, with a choice of MCP mode or a CLI plus skills mode.

Best for: Stopping Claude Code from using outdated APIs for fast-moving libraries.

Pricing: The README offers a free API key from the Context7 dashboard for higher rate limits. The project is MIT licensed. See our Context7 connect guide.

Pros

• Only two tools, so little context overhead.
• Free key option.
• Covers many libraries in one server.

Cons

• Docs are pulled from outside sources, so check important answers.
• The setup CLI needs Node.js 22.13 or newer.

Checked against: Context7 on GitHub on 28 October 2026.

3. Playwright MCP: drive a real browser

What it is. Playwright MCP from Microsoft "provides browser automation capabilities using Playwright." It works from structured accessibility snapshots instead of screenshots. The README's Claude Code command is claude mcp add playwright npx @playwright/mcp@latest.

The README is candid: "If you are using a coding agent, you might benefit from using the CLI+SKILLS instead." It positions MCP for exploratory automation and self-healing tests.

Best for: Checking that a page Claude Code just built actually works.

Pricing: Free and open source under Apache-2.0. See our Playwright connect guide.

Pros

• Free, local and official.
• Text snapshots are cheaper than images.
• Good for testing forms and flows.

Cons

• Microsoft itself suggests the CLI for many coding-agent tasks.
• Gives the AI whatever the browser can see.

Checked against: Playwright MCP on GitHub on 28 October 2026.

4. Sentry MCP: errors and performance

What it is. Sentry's MCP server covers error search, performance analysis, issue triage and docs. The Claude Code command is claude mcp add --transport http sentry https://mcp.sentry.dev/mcp/{organizationSlug}/{projectSlug}.

"All connections use OAuth." Sentry says "Scoping to a project is recommended when possible."

Best for: Fixing production bugs with the real stack trace in front of Claude.

Pricing: The page lists no MCP price; you need a Sentry account. See our Sentry connect guide.

Pros

• Scope to one org or project in the URL.
• OAuth, no tokens in config files.
• Covers issues and performance.

Cons

• Experimental tools need an extra flag.
• No price information on the MCP page.

Checked against: Sentry MCP on 28 October 2026.

5. Vercel MCP: deployments and logs

What it is. Vercel MCP is "a remote MCP with OAuth" at https://mcp.vercel.com. Claude Code is first on its list of approved clients. You add it with claude mcp add --transport http vercel https://mcp.vercel.com and sign in with /mcp.

It searches Vercel docs, manages projects and deployments, reads deployment logs and queries analytics.

Best for: Next.js and front-end teams that deploy on Vercel.

Pricing: "Vercel MCP is available on all plans," subject to Vercel's AI Product Terms. See our Vercel connect guide.

Pros

• Only reviewed clients can connect.
• Explicit consent per client blocks confused deputy attacks.
• Logs and analytics in one place.

Cons

• Same access as your Vercel account.
• Deploy and purchase tools mean real changes are possible.

Checked against: Vercel MCP docs on 28 October 2026.

6. Supabase MCP: Postgres with guardrails

What it is. Supabase's MCP guide gives claude mcp add --scope project --transport http supabase "https://mcp.supabase.com/mcp". URL flags let you lock it down: read_only=true, project_ref=<id> for one project, and features= for chosen tool groups.

The guide warns, "Connecting an LLM to your Supabase projects carries security risks," mainly prompt injection hidden in your data.

Best for: App builders on Supabase who want Claude Code to read schemas and write migrations in development.

Pricing: No separate MCP price. Branching needs a paid plan, and creating projects or branches shows a cost confirmation first. See our Supabase connect guide.

Pros

• Read-only and single-project modes.
• Cost confirmation before paid actions.
• Project scope keeps it in .mcp.json for the team.

Cons

• Not meant for customers or end users.
• Risky on production data without the guardrails.

Checked against: Supabase MCP guide on 28 October 2026.

7. Neon MCP: branchable Postgres

What it is. Neon's MCP docs describe a hosted server at https://mcp.neon.tech/mcp. npx neon@latest init sets up the server, agent skills and a linked project, and Neon lists Claude Code among agents that support its plugin.

Adding ?readonly=true keeps SELECT queries and schema inspection but disables writes. Neon says, "Never connect MCP agents to production databases."

Best for: Testing schema changes on a throwaway database branch.

Pricing: The docs list no MCP price; normal Neon usage applies. See our Neon connect guide.

Pros

• Branch tools suit safe experiments.
• OAuth or API key.
• Read-only by URL or OAuth scope.

Cons

• Recommended for development and testing only.
• The /sse endpoint is deprecated.

Checked against: Neon MCP docs on 28 October 2026.

8. Figma MCP: designs to code

What it is. Figma's remote server docs give two Claude Code options: claude plugin install figma@claude-plugins-official, or claude mcp add --transport http figma https://mcp.figma.com/mcp. Figma prefers the plugin because it bundles skills too.

"Only clients listed in the Figma MCP catalog" can connect. You paste a "Copy link to selection" URL into your prompt to point Claude at a frame.

Best for: Front-end developers building from Figma files.

Pricing: Figma's limits page gives View and Collab seats 20 tool calls per month. Full and Dev seats on Professional get 200 per day, and Enterprise gets 600 per day. Some tools use Figma AI credits. See our Figma connect guide.

Pros

• Official plugin for Claude Code.
• Link-to-selection keeps prompts precise.
• No desktop app needed for the remote server.

Cons

• Tight limits on View and Collab seats.
• Some tools spend AI credits.

Checked against: Figma remote server docs on 28 October 2026.

9. Stripe MCP: payments code and data

What it is. Stripe's MCP docs give claude mcp add --transport http stripe https://mcp.stripe.com/ and OAuth through /mcp. You can also use an agent API key from an environment variable in .mcp.json.

Stripe adds its own check: refunds and outbound payments need human confirmation. Starting in December 2026, the server stops accepting full-access secret keys and restricted keys without the Agent tag.

Best for: Developers building checkout, billing or subscription features.

Pricing: No separate MCP price. Normal Stripe fees apply to real transactions. See our Stripe connect guide.

Pros

• Sandbox and live access set separately.
• Built-in human confirmation for money movement.
• Docs search tool helps with integration code.

Cons

• Key-based setups must move to agent keys by 30 November 2026.
• Broad API write tool if you grant it.

Checked against: Stripe MCP docs on 28 October 2026.

10. Cloudflare MCP: the whole platform

What it is. Cloudflare's MCP page describes https://mcp.cloudflare.com/mcp as access to "over 2,500 endpoints" plus docs. It uses Code Mode: two tools, search() and execute(), with the model writing code that runs in a sandbox.

Cloudflare estimates this uses about 1,000 tokens versus roughly 1,170,000 for full schemas. Separate servers cover docs, containers and Browser Run.

Best for: Teams on Workers, Pages, DNS or other Cloudflare products.

Pricing: The docs page does not mention pricing. Sign-in is OAuth, or an API token for automation. See our Cloudflare connect guide.

Pros

• One server for the whole API.
• Very low context cost.
• OAuth lets you choose permissions.

Cons

• Code execution is powerful, so scope tokens carefully.
• No pricing on the MCP page.

Checked against: Cloudflare MCP servers docs on 28 October 2026.

Setting up servers safely in Claude Code

Use project scope for servers the whole team needs, so they sit in .mcp.json in version control. Keep personal tokens in local or user scope, stored in ~/.claude.json. Claude Code's docs say it "prompts for approval in interactive sessions before using project-scoped servers."

Watch tool output size too. Claude Code warns above 10,000 tokens and caps output at 25,000 by default, adjustable with MAX_MCP_OUTPUT_TOKENS. Servers with read-only or scoped modes, like Supabase and Sentry, also keep output smaller.

Next step

If you want Claude Code set up with the right servers, permissions and team conventions, see our Claude Code consulting. Browse more lists in the MCP Guide's best-of section.

Frequently asked questions