--- slug: "best-mcp-servers-for-coding-agents" title: "MCP servers for coding agents, picked by job" description: "MCP servers for Claude Code, Cursor and Codex, grouped by job, with the install command from each project's docs, what each needs, license and GitHub data." published: "2026-10-15" updated: "2026-10-15" signals_fetched_at: "2026-10-15T05:02:49.930Z" tools: ["serena", "context7", "github-mcp-server", "chrome-devtools-mcp", "playwright-mcp", "apify-mcp", "firecrawl-mcp", "exa-mcp", "supabase-mcp", "stripe-agent-toolkit", "figma-context-mcp", "sentry-mcp"] page: "https://agenttooldrop.com/guides/best-mcp-servers-for-coding-agents" --- # MCP servers for coding agents, picked by job Updated 2026-10-15. Page: https://agenttooldrop.com/guides/best-mcp-servers-for-coding-agents An MCP server gives a coding agent something it does not have built in: current library docs, a real browser, your GitHub issues, your database or your error tracker. These are the MCP servers in our catalog that are documented for coding agents such as Claude Code, Cursor and Codex, grouped by the job they do. Every row carries the install command or config block copied from the project's own docs. Best here means three things, and nothing more: each server is verified against its own repo and docs, each is documented for at least one coding agent, and within each group the new and rising servers come first, by new and rising score, then the standards by name. We did not benchmark them. ## Keep the set small Every server you connect adds tools the model has to choose between, and describing those tools to the model takes context. Microsoft makes this point in the Playwright MCP README, where it suggests its CLI with skills for coding agents that care about context size. Start with the two or three servers that match work you do every week, and add another when a task needs it. Several servers here act on real accounts: GitHub, Supabase and Stripe. Their own docs recommend scoped projects, read-only modes, tool approval or restricted keys, and each row's skip-it-if notes say which. ## Installing in your client A `claude mcp add` command is for Claude Code. An `mcpServers` JSON block goes in the MCP configuration of clients that read that format. Each tool page links to the docs page the snippet came from, which covers the other clients that project supports. ## At a glance | Tool | You need | License | Pricing | Install | Stars | Last push | | --- | --- | --- | --- | --- | --- | --- | | [Serena](https://agenttooldrop.com/tools/serena) | uv (the snippet pins Python 3.13) | GPL-3.0-or-later | Open source | Install with uv | 30,271 | 2026-10-14 | | [Context7](https://agenttooldrop.com/tools/context7) | Nothing to start; an API key for higher rate limits | MIT | Free tier + paid | Run the setup CLI | 63,053 | 2026-10-14 | | [GitHub MCP Server](https://agenttooldrop.com/tools/github-mcp-server) | A GitHub account; a personal access token in some clients | MIT | Open source | Add the remote server to Claude Code | 33,623 | 2026-10-14 | | [Chrome DevTools MCP](https://agenttooldrop.com/tools/chrome-devtools-mcp) | Chrome or Chrome for Testing; Node.js (npx) | Apache-2.0 | Open source | Add to your MCP client config | 53,560 | 2026-10-15 | | [Playwright MCP](https://agenttooldrop.com/tools/playwright-mcp) | Node.js (runs with npx) | Apache-2.0 | Open source | Add to Claude Code | 38,158 | 2026-10-08 | | [Apify MCP Server](https://agenttooldrop.com/tools/apify-mcp) | An Apify account, token or agentic payment | MIT | Free tier + paid | Add the hosted server (OAuth) | 11,273 | 2026-10-14 | | [Firecrawl MCP Server](https://agenttooldrop.com/tools/firecrawl-mcp) | A Firecrawl account or API key for crawl, map and agent tools | MIT | Open source | Add the hosted server (OAuth sign-in) | 7,607 | 2026-10-15 | | [Exa MCP Server](https://agenttooldrop.com/tools/exa-mcp) | Nothing for light, rate-limited use; OAuth or an Exa API key for Exa Agent | MIT | Open source | Install the Claude Code plugin | 5,130 | 2026-10-07 | | [Supabase MCP Server](https://agenttooldrop.com/tools/supabase-mcp) | A Supabase project | Apache-2.0 | Open source | Add the hosted server to your MCP config | 2,952 | 2026-10-14 | | [Stripe MCP and Agent Toolkit](https://agenttooldrop.com/tools/stripe-agent-toolkit) | A Stripe account; restricted API keys advised | MIT | Open source | Add to Claude Code | 1,943 | 2026-10-15 | | [Framelink MCP for Figma](https://agenttooldrop.com/tools/figma-context-mcp) | A Figma personal access token | MIT | Open source | Add to your MCP client config (macOS / Linux) | 15,988 | 2026-10-14 | | [Sentry MCP Server](https://agenttooldrop.com/tools/sentry-mcp) | A Sentry organization and project | FSL-1.1-Apache-2.0 | Free | Add to Claude Code | 928 | 2026-10-15 | ## Library docs and code context ### Serena Gives a coding agent symbol-level code search, editing and refactoring tools backed by language servers Pick it if your agent spends many steps on grep and line-based edits in a large codebase and should look up symbols and references instead. License: GPL-3.0-or-later (the app, since v2; v1.7.0 and earlier MIT); SolidLSP library MIT. Pricing: Open source. GitHub: 30,271 stars, last push 2026-10-14. Skip it if: Your projects are small and the agent's built-in search and edit tools already cope. It reports anonymous usage data on startup unless SERENA_USAGE_REPORTING is set to false, and its docs assume a trusted machine and repository and recommend a sandbox. The app moved from MIT to GPL-3.0-or-later in v2. Install with uv: ```sh uv tool install -p 3.13 serena-agent ``` Install source: https://github.com/oraios/serena#readme. Entry: https://agenttooldrop.com/tools/serena.md ### Context7 Pulls current, version-specific library docs and code examples into your agent's context Pick it if your agent keeps writing code against outdated or nonexistent APIs for the libraries you depend on. Requests go to the hosted Context7 API. License: MIT (MCP server and CLI); the API backend, parsing and crawling engines are private. Pricing: Free tier + paid. GitHub: 63,053 stars, last push 2026-10-14. Skip it if: You need a fully self-hosted setup: only the MCP server and CLI are in the repo, and requests go to the hosted Context7 API, with low rate limits unless you add an API key. Library docs are community-contributed and Context7 does not guarantee their accuracy. Run the setup CLI: ```sh npx ctx7 setup ``` Install source: https://github.com/upstash/context7#readme. Entry: https://agenttooldrop.com/tools/context7.md ### GitHub MCP Server Connects agents to GitHub repos, issues, pull requests, Actions runs and security alerts Pick it if your agent needs to read or act on GitHub repositories, issues, pull requests or Actions runs as part of its task. License: MIT. Pricing: Open source. GitHub: 33,623 stars, last push 2026-10-14. Skip it if: You do not use GitHub. Remote OAuth depends on each host configuring a GitHub App or OAuth App, so some clients need a personal access token, and GitHub Enterprise Server needs extra setup. Add the remote server to Claude Code: ```sh claude mcp add-json github '{"type":"http","url":"https://api.githubcopilot.com/mcp","headers":{"Authorization":"Bearer YOUR_GITHUB_PAT"}}' ``` Install source: https://github.com/github/github-mcp-server/blob/main/docs/installation-guides/install-claude.md. Entry: https://agenttooldrop.com/tools/github-mcp-server.md ## A real browser ### Chrome DevTools MCP Lets a coding agent control and inspect a live Chrome browser through DevTools Pick it if you want the agent to check its changes in a real Chrome instance and read console, network and performance data. License: Apache-2.0. Pricing: Open source. GitHub: 53,560 stars, last push 2026-10-15. Skip it if: You need Firefox or WebKit; only Chrome and Chrome for Testing are officially supported. Usage statistics go to Google by default (opt out with --no-usage-statistics), and performance tools may send trace URLs to the CrUX API unless you pass --no-performance-crux. Add to your MCP client config: ```json { "mcpServers": { "chrome-devtools": { "command": "npx", "args": [ "-y", "chrome-devtools-mcp@latest" ] } } } ``` Install source: https://github.com/ChromeDevTools/chrome-devtools-mcp#readme. Entry: https://agenttooldrop.com/tools/chrome-devtools-mcp.md ### Playwright MCP Lets an agent drive a real browser with Playwright using accessibility snapshots, not screenshots Pick it if your agent needs to navigate, fill forms and inspect pages over many steps in one browser session. License: Apache-2.0. Pricing: Open source. GitHub: 38,158 stars, last push 2026-10-08. Skip it if: You use a coding agent and care about context size; Microsoft suggests Playwright CLI with skills for that. It is not a security boundary, so keep it away from sessions or data you would not hand to the model. Add to Claude Code: ```sh claude mcp add playwright npx @playwright/mcp@latest ``` Install source: https://github.com/microsoft/playwright-mcp#readme. Entry: https://agenttooldrop.com/tools/playwright-mcp.md ## Web search and scraping ### Apify MCP Server Lets an agent search Apify Store and run its ready-made scrapers and crawlers as MCP tools Pick it if an existing Apify Actor already scrapes the site you need and paying per run beats maintaining a scraper. License: MIT for the server code; Actor runs are billed as Apify platform usage. Pricing: Free tier + paid. GitHub: 11,273 stars, last push 2026-10-14. Skip it if: You want a free or offline tool: running Actors needs an Apify account, token or agentic payment and uses paid platform usage beyond the Free plan's monthly allowance. Telemetry on tool calls is on by default; the README documents how to turn it off. Add the hosted server (OAuth): ```json { "mcpServers": { "apify": { "url": "https://mcp.apify.com" } } } ``` Install source: https://docs.apify.com/mcp. Entry: https://agenttooldrop.com/tools/apify-mcp.md ### Firecrawl MCP Server Lets agents search, scrape, crawl and interact with web pages through Firecrawl, returning markdown or JSON Pick it if your agent needs content from known URLs, a whole site or a search query as markdown or JSON. License: MIT. Pricing: Open source. GitHub: 7,607 stars, last push 2026-10-15. Skip it if: You want to drive a browser session step by step from your own client: each interact call runs to completion and returns control. Crawl, map and agent tools need a Firecrawl account or API key. Add the hosted server (OAuth sign-in): ```json { "mcpServers": { "firecrawl": { "type": "http", "url": "https://mcp.firecrawl.dev/v2/mcp-oauth" } } } ``` Install source: https://docs.firecrawl.dev/mcp-server. Entry: https://agenttooldrop.com/tools/firecrawl-mcp.md ### Exa MCP Server Gives agents Exa web search and page fetching as MCP tools, with optional multi-step research runs Pick it if your agent needs web search and pages as clean markdown from a hosted server, with nothing to install. License: MIT. Pricing: Open source. GitHub: 5,130 stars, last push 2026-10-07. Skip it if: You need more than light anonymous use: the keyless server is rate-limited, and Exa Agent requires OAuth or an Exa API key. Install the Claude Code plugin: ```sh claude plugin install exa@claude-plugins-official ``` Install source: https://github.com/exa-labs/exa-mcp-server#readme. Entry: https://agenttooldrop.com/tools/exa-mcp.md ## Your database, errors, designs and payments ### Supabase MCP Server Lets an AI assistant work with Supabase projects: tables, migrations, SQL, logs and Edge Functions Pick it if your app runs on Supabase and the agent should read schemas, run SQL, apply migrations or check logs. Supabase recommends project scoping and read-only mode. License: Apache-2.0. Pricing: Open source. GitHub: 2,952 stars, last push 2026-10-14. Skip it if: You do not use Supabase, or you cannot keep it away from production data: Supabase warns about prompt injection, recommends project scoping and read-only mode, and says not to give the server to your end users. The local CLI and self-hosted setups get only a subset of tools and no OAuth. Add the hosted server to your MCP config: ```json { "mcpServers": { "supabase": { "type": "http", "url": "https://mcp.supabase.com/mcp" } } } ``` Install source: https://github.com/supabase/mcp#readme. Entry: https://agenttooldrop.com/tools/supabase-mcp.md ### Stripe MCP and Agent Toolkit Lets agents call the Stripe API and search Stripe docs through a hosted MCP server or framework toolkits Pick it if you build on Stripe. Write tools act on a real Stripe account, so Stripe advises tool approval and restricted API keys. License: MIT. Pricing: Open source. GitHub: 1,943 stars, last push 2026-10-15. Skip it if: You do not use Stripe. Write tools act on a real Stripe account, so Stripe advises turning on tool approval in your client and using restricted API keys to limit what the toolkit can do. Add to Claude Code: ```sh claude mcp add --transport http stripe https://mcp.stripe.com/ ``` Install source: https://docs.stripe.com/mcp. Entry: https://agenttooldrop.com/tools/stripe-agent-toolkit.md ### Framelink MCP for Figma Gives a coding agent the layout and styling data behind a Figma file, frame or group link Pick it if you build UI from Figma designs and want the agent to read the file data instead of a screenshot. License: MIT. Pricing: Open source. GitHub: 15,988 stars, last push 2026-10-14. Skip it if: You do not use Figma or cannot create a Figma personal access token. The README says it is designed specifically for Cursor, though the same config works in other MCP clients. Add to your MCP client config (macOS / Linux): ```json { "mcpServers": { "Framelink MCP for Figma": { "command": "npx", "args": [ "-y", "figma-developer-mcp", "--figma-api-key=YOUR-KEY", "--stdio" ] } } } ``` Install source: https://github.com/GLips/Figma-Context-MCP#readme. Entry: https://agenttooldrop.com/tools/figma-context-mcp.md ### Sentry MCP Server Gives coding agents access to Sentry errors, issues, traces and performance data over MCP Pick it if your app reports to Sentry and the agent should pull issue details, errors and traces while it works on a fix. License: FSL-1.1-Apache-2.0 (source-available; each version converts to Apache-2.0 two years after release). Pricing: Free. GitHub: 928 stars, last push 2026-10-15. Skip it if: You do not use Sentry. Sentry says it is not a general-purpose server for all Sentry functionality, and the stdio transport is still a work in progress. Add to Claude Code: ```sh claude mcp add --transport http sentry https://mcp.sentry.dev/mcp/{organizationSlug}/{projectSlug} ``` Install source: https://mcp.sentry.dev. Entry: https://agenttooldrop.com/tools/sentry-mcp.md ## Newer options worth a look New and rising tools from the catalog that this guide does not compare yet, by new and rising score. Younger and less proven than the picks above. - [agent-memory](https://agenttooldrop.com/tools/agent-memory.md): Gives coding agents a shared long-term memory kept as plain Markdown files, with local ranked recall. Skip it if: You want a packaged install: there is no PyPI release yet, so it installs from a git checkout with uv, and the repo was created on 2026-09-01. Setup writes lifecycle hooks into your agent's settings and runs the host agent's CLI headlessly to distil and consolidate memories. - [Hex-Rays IDA MCP](https://agenttooldrop.com/tools/ida-mcp.md): Connects an agent to IDA to open, analyse and script binaries, headless or in the IDA GUI. Skip it if: You do not have IDA 9.4 or later with idalib and Python 3.11+. The agent runs Python inside IDA, and sessions are logged locally with links to agent transcripts; Hex-Rays recommends disabling other IDA MCP servers. - [Browser Harness](https://agenttooldrop.com/tools/browser-harness.md): Lets Claude Code, Codex or Cursor drive your own Chrome over CDP, logged in as you. Skip it if: You do not want an agent acting in your real, logged-in browser profile: setup turns on Chrome remote debugging, and the agent acts as you on the sites you use. Optional local recordings save screenshots and action traces that may include sensitive page content. - [Godot AI](https://agenttooldrop.com/tools/godot-ai.md): Connects Claude Code, Codex and other MCP clients to a live Godot editor to build scenes, nodes and scripts. Skip it if: You work mainly in C#, which is text-only (it does not build .NET or report compiler errors), or you are on Godot older than 4.7. Usage telemetry is on unless GODOT_AI_DISABLE_TELEMETRY=true is set, and the agent gets write access to the open project. ## How we chose Picked from the 95 tools in our catalog, each verified against its own repository and docs; 12 made this guide. Within each group, new and rising tools come first, sorted by the new and rising score (star velocity, recent pushes and releases, repo age and an editorial weight), then the standards, the well-known tools, by name. It is not a benchmark. The number beside a tool is its rank in the full new and rising ledger, not within this guide; standards are not ranked. Install commands are copied from each project's docs; every tool page links to the page they came from. GitHub signals (stars, last push, latest release) were fetched on 2026-10-15. Left out on purpose: the Fetch and Filesystem reference servers, because most coding agents already fetch pages and edit files on their own and the servers repo calls them reference implementations, not production-ready servers; and the Model Context Protocol entry itself, which is for building servers and clients.