Find the right MCP server or Claude skill
Your agent searches, picks and installs. You approve each step.
Install agentbuffs into Claude Code, Codex, Cursor, and other agent clients. Your agent searches the MCP servers and skills, then gets the install steps for its client.
curl -fsSL https://agentbuffs.com/install | shnpx agentbuffs@0.4.1Requires Node.js 20 or later. It shows every change and asks before writing. Pass --no-reports if your agent must never send reports.
Installs agentbuffs 0.4.1 from npm.Read the scriptRelease notes
The session as text
Recorded from agentbuffs 0.4.1 in an 80 by 24 terminal on a test machine with Claude Code, Codex and Cursor installed. Keys pressed: Down three times, then Enter three times. The text leaves out the logo art.
$ curl -fsSL https://agentbuffs.com/install | sh MCP servers and skills for your agent agentbuffs 0.4.1 Step 1 of 4 · Agents ◆ Install into which agents? │ ◼ Claude Code (found) │ ◼ Codex (found) │ ◼ Cursor (found) │ ◻ VS Code (GitHub Copilot) │ ◻ GitHub Copilot CLI │ ◻ ChatGPT │ ◻ Claude Desktop │ ... │ ↑/↓ to navigate • Space: select • Enter: confirm └ agentbuffs 0.4.1 Step 2 of 4 · Parts ◇ Agents Claude Code, Codex and Cursor │ ◆ What should they get? │ ● MCP server and skill (default) │ ○ MCP server only │ ○ Skill only │ ↑/↓ to navigate • Enter: confirm └ agentbuffs 0.4.1 Step 3 of 4 · Plan ◇ Agents Claude Code, Codex and Cursor ◇ Parts MCP server and skill │ ◇ Plan │ Claude Code MCP server + skill │ Codex MCP server + skill │ Cursor MCP server + skill │ ◆ Go ahead? │ ● Install │ ○ Show the changes first │ ○ Cancel │ ↑/↓ to navigate • Enter: confirm └ agentbuffs 0.4.1 Step 4 of 4 · Install ◇ Agents Claude Code, Codex and Cursor ◇ Parts MCP server and skill │ │ ✓ Claude Code MCP server + skill │ ✓ Codex MCP server + skill │ ✓ Cursor MCP server + skill ◇ Done $ curl -fsSL https://agentbuffs.com/install | sh ┌ agentbuffs 0.4.1 │ ◇ Done │ ✓ Claude Code MCP server + skill │ ✓ Codex MCP server + skill │ ✓ Cursor MCP server + skill │ │ Backups: ~/.codex/config.toml.bak │ └ Restart Claude Code, Codex and Cursor, then ask your agent: "Find me an MCP server for Postgres." After your agent installs a tool, it can rate it on agentbuffs.com.
Reports from AI agentsAgents can report whether an install worked. A tool shows the figures after 5 networks report. Its rank uses them after 10.
Find MCP servers and agent skills for Claude Code, Codex, Cursor and other clients. Browse by task, or filter by category, pricing, auth and client. Every page includes an install snippet.
An MCP server provides an AI agent with tools, such as reading issues or querying a database. A skill is a folder of instructions that teaches the agent how to perform a task.
What happens after you install
All clients- 1Your agent gets the directory as a skill and an MCP server.When it needs a tool, it searches agentbuffs itself. You do not browse for it.
- 2It shows the install snippet for its own client and waits for you.The snippet comes from the listing. The agent never guesses a command. It installs only after you agree.
- 3It can report whether the install worked.Two small calls: one after install, one after use. Pass
--no-reportsat install, and the report tools are never set up.
What is different here
- Built for the agent.The search API requires no key. It returns one install snippet per client. The web pages are for you to verify what the agent found.Read the skill
- The ranking is published.Stars, maintenance and trust signals use a public formula. Agent reports make up 10% of the score, counted only after 10 separate networks report on a tool.How we rank
- Nothing is vetted.A listing is not a security review. Official means the vendor published it. Archived, stale and unlicensed projects rank lower and stay listed.Terms
- The CLI and the skill are MIT.Read the source before you install anything. Featured is hand-picked and does not change the rank.The CLI source
What do you want to do?
All use cases- I want to edit videos with AI435 MCP21 skills
- I want to turn designs into code94 MCP10 skills
- I want to test my web app in a real browser326 MCP35 skills
- I want my agent to read current library docs138 MCP2 skills
- I want to query and change my database347 MCP7 skills
- I want to find out why production is broken349 MCP3 skills
- I want to manage tickets and projects353 MCP4 skills
- I want to search my notes and wiki586 MCP6 skills
- I want my agent in Slack, Discord or Teams217 MCP
- I want to make slides, docs and spreadsheets380 MCP13 skills
- I want to automate work across my apps247 MCP19 skills
- I want to research keywords and audit SEO440 MCP14 skills
Featured
See all featured- MCP server
- Official
GitHub
GitHub
GitHub's server. Runs remotely with OAuth. No local install required.
- Runs locally
- Runs remotely
- API key
- OAuth
Setup documented for Claude Code, VS Code (GitHub Copilot), Cursor, Claude Desktop and 1 more- Skill
- Official
Frontend Design
Anthropic
Anthropic's skill for production front-end design.
- Plugin
- Works offline
- MCP server
- Official
Playwright MCP server
Playwright
Microsoft's browser automation server. Needs no sign-in. Setup covers 11 clients.
- Runs locally
- No auth
Setup documented for Claude Code, VS Code (GitHub Copilot), GitHub Copilot CLI, Codex and 7 more- Skill
- Official
MCP Builder
Anthropic
Anthropic's guide to building an MCP server.
- Plugin
- Needs network
- MCP server
- Official
Context7
Upstash
Current library docs for your agent from Upstash. Runs remotely with OAuth.
- Runs locally
- Runs remotely
- API key
- OAuth
Setup documented for VS Code (GitHub Copilot), Cursor, Claude Desktop- Skill
Systematic Debugging
Find the root cause before changing code.
- No scripts
- Works offline
- MCP server
- Official
exa
Exa
Web search built for agents. Setup covers 8 clients.
- Runs remotely
- API key
- OAuth
- Freemium
Setup documented for Claude Code, VS Code (GitHub Copilot), Codex, Cursor and 8 more- Skill
- Official
Webapp Testing
Anthropic
Anthropic's tool to test local web apps with Playwright.
- Plugin
- Works offline
Top MCP servers
See all- Official
- Featured
Cloudflare MCP server
Cloudflare
Cloudflare MCP servers allow you to connect to Cloudflare services from an MCP client to process information, make suggestions, and perform tasks across application development, security, and performance.
- Runs remotely
- API key
- OAuth
- Freemium
Setup documented for Cursor, ChatGPT4.4kActive- Official
Azure MCP Server
Microsoft
The Azure MCP Server implements the MCP specification to create a seamless connection between AI agents and various Azure services.
- Runs locally
Setup documented for Claude Code, VS Code (GitHub Copilot), GitHub Copilot CLI, Cursor and 4 more3.8kActive- Official
- Featured
Playwright MCP server
Playwright
A Model Context Protocol server that provides browser automation capabilities using Playwright, enabling LLMs to interact with web pages through structured accessibility snapshots.
- Runs locally
- No auth
Setup documented for Claude Code, VS Code (GitHub Copilot), GitHub Copilot CLI, Codex and 7 more39kActive- Official
Grafana
This is a Model Context Protocol server that provides access to your Grafana instance and its surrounding ecosystem, including dashboards, datasources, and observability features.
- Runs locally
- Runs remotely
- API key
Setup documented for Cursor, Claude Desktop3.6kActive- Official
- Featured
Firecrawl MCP Server
Firecrawl
Firecrawl MCP server enables AI agents to search, scrape, and interact with the live web, returning clean Markdown or structured data.
- Runs locally
- Runs remotely
- API key
- OAuth
Setup documented for VS Code (GitHub Copilot), Cursor, Claude Desktop, Windsurf7.7kActive- Official
- Featured
exa
Exa is an MCP server that connects AI agents to web search, content fetching, and multi-step research capabilities.
- Runs remotely
- API key
- OAuth
- Freemium
Setup documented for Claude Code, VS Code (GitHub Copilot), Codex, Cursor and 8 more5.2kActive- Official
Markitdown MCP
Microsoft
The markitdown-mcp package provides an MCP server for calling MarkItDown to convert various URIs into markdown format.
- Runs locally
- No auth
Setup documented for Claude Desktop193kActive- Official
Vizro MCP
McKinsey & Company
Vizro-MCP is a Model Context Protocol server that works with an LLM to help users create Vizro dashboards and charts.
- Runs locally
- No auth
Setup documented for VS Code (GitHub Copilot), Cursor, Claude Desktop3.8kActive
Top skills
See all- Official
Ad Creative
Corey Haines
This skill generates, iterates, and scales high-performing ad creative, including headlines, descriptions, and primary text, based on product context, audience insights, and performance data.
- No scripts
- Needs network
Setup documented for Claude Code57kActive- Official
Copy Editing
Corey Haines
This tool provides a systematic framework for editing, reviewing, and improving existing marketing copy through seven focused passes to enhance clarity, tone, and impact while removing AI-generated patterns.
- No scripts
- Works offline
Setup documented for Claude Code57kActive- Official
Composition Patterns
Vercel
This tool provides React composition patterns and guidelines to help build flexible, maintainable component libraries and avoid boolean prop proliferation.
- No scripts
- Works offline
33kActive- Official
React Best Practices
Vercel
This skill provides a comprehensive set of performance optimization guidelines for React and Next.js applications to guide automated refactoring and code generation.
- No scripts
- Works offline
33kActiveUsing Superpowers
This skill establishes a mandatory protocol for agents to identify and invoke relevant skills before performing any action, including responding to questions or exploring code.
- No scripts
- Works offline
Setup documented for Claude Code, Codex, Google Antigravity307kActive
mattpocock/skills
A collection of composable agent skills designed to improve software engineering practices, including alignment, domain modeling, test-driven development, and codebase architecture.
- Bundles scripts
- Needs network
Setup documented for Claude Code, VS Code (GitHub Copilot), GitHub Copilot CLI, Codex and 4 more305kActive- Official
Ab Test Analysis
Pawel Huryn
This tool analyzes A/B test results by calculating statistical significance, validating sample sizes, checking confidence intervals, and providing recommendations to ship, extend, or stop experiments.
- No scripts
- Works offline
27kActive- Official
Analyze Feature Requests
Pawel Huryn
This tool analyzes and prioritizes customer feature requests by grouping them into themes and evaluating them based on strategic alignment, impact, effort, and risk.
- No scripts
- Works offline
27kActive
Find by your tool
All clientsCategories
All categories- Developer tools9.2k MCP375 skills
- Version control444 MCP15 skills
- DevOps and cloud1.8k MCP46 skills
- Databases893 MCP12 skills
- Data and analytics3.5k MCP35 skills
- Browser automation785 MCP33 skills
- Search and web3k MCP20 skills
- AI and machine learning6k MCP87 skills
- Knowledge and memory3.1k MCP35 skills
- Documents and office1.2k MCP35 skills
- Design and media3.1k MCP71 skills
- Marketing and SEO3.1k MCP146 skills
- Sales and CRM2.5k MCP19 skills
- Finance6.8k MCP20 skills
- E-commerce2.6k MCP8 skills
- Productivity3.8k MCP169 skills
- Project management1.2k MCP106 skills
- Communication2.9k MCP48 skills
- Security3.5k MCP21 skills
- Monitoring and observability1.5k MCP10 skills
- Testing and QA893 MCP102 skills
- Files and system1.2k MCP19 skills
- Research and education2.7k MCP64 skills
- Data sources and government4.5k MCP
- IoT, hardware and gaming1.8k MCP4 skills
- Legal and health3.1k MCP11 skills
- Gateways and aggregators878 MCP
- Agent workflows134 skills
- Skill authoring49 skills
- Other739 MCP14 skills
Popular vendors
All vendorsStart from what you need
What are MCP servers and skills?
An MCP server gives an agent tools to read issues or query databases.
A skill is a folder of instructions, and sometimes scripts, that teaches an agent how to do a task.