Resources

Short, practical guides for choosing MCP servers and agent skills, plus links to the official documentation.

How to read each listing

Type: MCP server (a connector that gives an agent tools), agent skill (instructions and scripts in a SKILL.md folder, or a prompt pack) or agent tooling (frameworks, SDKs and utilities).
Official or community: official means published by the company behind the service, or by the MCP project or Anthropic. Community servers can be excellent, but check who maintains them.
Stars: how many people bookmarked the GitHub repository. A rough popularity signal, not a quality or safety rating. For folders inside a bigger repository, stars are for the whole repository.
Last update: the latest commit on the repository, or on the listing's folder for monorepos. Fast-moving protocols mean stale servers break.
Licence: the SPDX licence from the repository. Hosted servers with no public code are shown as proprietary.
Install: from the MCP Registry entry or the README. “Remote” means a URL you add to your client; npm, PyPI and Docker hints name the package to run.

Prefer official servers for business data: a vendor's own server follows its API and permission model and gets fixes when the API changes.
Grant the least access: use read-only modes, restricted API keys and separate test accounts where they exist.
Watch for prompt injection: content an agent reads (web pages, issues, emails) can contain instructions. Be careful combining servers that read untrusted content with ones that can send data out.
Pin versions: pin the package version you tested rather than always pulling latest.
Scan your setup: tools such as Snyk Agent Scan check installed servers and skills for risky tool descriptions.

What a skill is: a folder with a SKILL.md file (name, description, instructions) plus optional scripts and reference files. The agent loads it only when the description matches the task.
Where they work: Claude Code, the Claude apps and API, and a growing number of other agents that support the Agent Skills format, including Codex.
Read before installing: skills can include scripts the agent will run, so read them as you would any code.
Check the licence: Anthropic's document skills are source-available rather than open source; most other skills here use permissive licences.

Model Context Protocol documentation ↗The specification, concepts and guides for building servers and clients.
Official MCP Registry ↗The community registry of published MCP servers, with an open API.
modelcontextprotocol/servers ↗The MCP project's reference servers.
Agent Skills (anthropics/skills) ↗Anthropic's skills repository, with the format spec and a template.

Know a guide that should be here? Tell us.