MCP servers directory

A curated list of MCP servers that work with Claude Code, Cursor, Windsurf, and other MCP-aware hosts. Grouped by what they do. AgentDrop is included where it fits (encrypted file transfer between agents), alongside the servers we use ourselves and the servers people actually search for.

What is an MCP server?

An MCP server is a small program that exposes tools an AI assistant can call over the Model Context Protocol. Host applications like Claude Code, Cursor, and Windsurf launch these servers on startup, register their tools, and route tool calls from the language model to whichever server claims the matching name. Each server extends the model's reach into a different domain: local files, browsers, databases, code repositories, agent communication.

For a longer answer with examples, see our pages on MCP server for file transfer and encrypted file transfer for AI agents.

Filesystem and local files

The category every MCP setup starts with. These servers let an LLM read, write, and search the local filesystem the host has access to.

@modelcontextprotocol/server-filesystem - the official Anthropic filesystem MCP server. Reads and writes within a user-allowed root path. Most hosts ship a config snippet for it.
@modelcontextprotocol/server-everything - the official reference / debugging server. Exposes every tool type at once. Use it to sanity-check a host before debugging your real server.

Code, repositories, version control

@modelcontextprotocol/server-github - official GitHub MCP server. Read/write issues, PRs, and repo files via the GitHub API. Handles PAT auth.
@modelcontextprotocol/server-git - local git operations without leaving the host. Works against repos already on disk.

Browser automation

Playwright MCP server (Microsoft) - drives a real browser for an LLM. Lets agents click, type, and read pages. The big one for web-task automation.
Puppeteer MCP server - similar shape, Chrome-only. Older, simpler.
@modelcontextprotocol/server-fetch - HTTP fetch as a tool. Lighter than Playwright, useful when you only need page text.

Databases

@modelcontextprotocol/server-postgres - read-only Postgres queries with schema introspection. Don't hand the model a write connection unless you want it to run UPDATE.
@modelcontextprotocol/server-sqlite - same shape, SQLite files.
Stripe MCP server - query Stripe data without a custom integration.

Cloud and infrastructure

Azure MCP server - Microsoft's official Azure tool surface for MCP. Manages resources, queries logs.
Cloudflare MCP server - Workers, KV, R2, DNS.
Google Drive MCP server - file access on Google Drive, useful for personal docs but not for agent-to-agent file transfer (no client-side encryption).

Agent-to-agent communication

The category most MCP directories miss. Most MCP servers connect a model to a backend tool. These connect a model to another agent running somewhere else.

AgentDrop - encrypted file transfer between AI agents, including agents running in different MCP hosts on different machines. Tools: send_file, check_inbox, download_transfer, list_sendable. Client-side X25519 + AES-256-GCM. Free tier covers 50 transfers and 50 MB files per month, no card required.
Reddit MCP server - read Reddit threads as if Reddit were another agent. One-direction only.
Slack MCP server - send and read Slack messages from an LLM. Not encrypted at the agent level (Slack workspace permissions still apply).

Memory and context

Matrix - persistent semantic memory for Claude Code. Embeddings + SQLite. Useful when you want the model to remember work across sessions.
mem0 MCP server - long-term memory with namespaces.

How to add any of these to your MCP host

Every MCP host accepts the same JSON config shape. The host launches each listed server on startup, asks it for its tools, and exposes those tools to the model. The pattern:

{
  "mcpServers": {
    "agentdrop": {
      "command": "npx",
      "args": ["-y", "agentdrop@latest"],
      "env": {
        "AGENTDROP_API_KEY": "agd_..."
      }
    },
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/me/projects"]
    }
  }
}

Drop this into ~/.claude/mcp.json for Claude Code, the equivalent settings panel for Cursor or Windsurf, and the model picks up both servers on next launch.

What this list is not

We're not ranking MCP servers by popularity or rating them out of ten. Quality of an MCP server depends on the use case: a filesystem server is right for local work, AgentDrop is right for cross-agent file exchange, an Azure server is right for cloud automation. Pick what matches the job, then add a second server when the first hits its scope.

We'll keep this list updated as new MCP servers land. If you maintain one we should include, email [email protected].

Where to go next

MCP server for file transfer between AI agents - the AgentDrop angle in detail
Encrypted file transfer for AI agents - cryptography deep dive

Or sign up free to try AgentDrop. 50 transfers and 50 MB files included on the free tier, no card required.