MCP servers list: a curated directory for AI agents
Model Context Protocol (MCP) is the open standard that lets an AI agent call real tools — your filesystem, your database, your APIs. Below are 10 servers worth wiring up first, grouped by what they do, with the exact package name and config shape for each. New to the protocol? Start with the MCP guide.
Filesystem & local files
Scoped read/write access so an agent can actually edit your project.
| Server | What it does | Transport |
|---|---|---|
| Filesystem@modelcontextprotocol/server-filesystem | Read, write and list files inside directories you explicitly allow. | stdio |
| Git@modelcontextprotocol/server-git | Status, diff, log, blame and branch operations on a local repository. | stdio |
Databases
Schema-aware querying, so the agent stops guessing your column names.
| Server | What it does | Transport |
|---|---|---|
| PostgreSQL@modelcontextprotocol/server-postgres | Inspect schema and run read-only SQL against a Postgres connection string. | stdio |
| SQLite@modelcontextprotocol/server-sqlite | Query and explore a local SQLite database file. | stdio |
Web & APIs
Let the agent fetch live context instead of hallucinating documentation.
| Server | What it does | Transport |
|---|---|---|
| Fetch@modelcontextprotocol/server-fetch | HTTP fetching with markdown conversion — good for docs and internal APIs. | stdio |
| Puppeteer@modelcontextprotocol/server-puppeteer | Headless browser automation: navigate, click, screenshot, scrape. | stdio |
Developer workflow
Issue trackers, memory and chat where the work actually happens.
| Server | What it does | Transport |
|---|---|---|
| GitHub@modelcontextprotocol/server-github | Issues, pull requests, code search and file operations on GitHub repos. | stdio |
| Slack@modelcontextprotocol/server-slack | Read channels and post messages from an agent workflow. | stdio |
| Memory@modelcontextprotocol/server-memory | Persistent knowledge-graph memory across sessions. | stdio |
Hosted / remote MCP
Servers you reach over HTTP instead of spawning locally.
| Server | What it does | Transport |
|---|---|---|
| MANOVIK MCPhttps://manovik.in/mcp | MANO 1.1 reasoning, coding, writing, translation and supply-chain scanning as MCP tools inside Claude, ChatGPT or Cursor. | http |
How to add one
Almost every MCP client reads the same config shape. A stdio server is spawned as a child process; a hosted server is reached over HTTP.
{
"mcpServers": {
"filesystem": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/you/projects"]
},
"manovik": {
"url": "https://manovik.in/mcp"
}
}
}Choosing safely
- Scope filesystem servers to a single project directory, never your home folder.
- Use read-only database credentials unless the agent genuinely needs writes.
- Treat an MCP server like a dependency: check who publishes it before letting it read your code.
Use MANOVIK as your MCP server
Point any MCP client at https://manovik.in/mcp and your assistant gains MANO 1.1 — planning, adversarial review and synthesis on every answer.