Claude Skills and MCP Servers
Explore source-verified skills and MCP servers with documented compatibility for Claude clients and agent workflows.
Selection guidance
Use these checks before opening source files, installing a skill, or connecting an MCP server.
How to choose
Choose by task fit first: a focused skill is useful for repeatable instructions, while an MCP server is appropriate when Claude needs live access to an external system. Prefer profiles with clear setup, source evidence, and current maintenance.
Compatibility checks
Distinguish Claude Code, Claude Desktop, and other Claude-compatible clients. Verify the exact client, transport, runtime, configuration shape, and installation scope rather than assuming one Claude setup works everywhere.
Security considerations
Review every requested credential and permission. Use read-only access where possible, isolate untrusted repositories and remote servers, and require confirmation before write or destructive actions.
Guides for setup and real workflows
Source-cited guides explain the format, configuration boundary, security checks, and repeatable ways to apply these resources.
What SKILL.md is
Understand the open Agent Skills format, required frontmatter, progressive loading, optional scripts and references, and the checks to make before using a skill.
Read guideAgent skills setupInstall agent skills
Install portable Agent Skills in Codex or Claude Code, choose project versus user scope, verify discovery, and update or remove a skill without losing reviewability.
Read guideArchitecture guideMCP vs. agent skills
Choose between MCP servers and Agent Skills by separating live capabilities and authorization from reusable instructions, workflow policy, templates, and examples.
Read guideEngineering workflowAgent skills for code review
Design or evaluate a code-review skill that reconstructs intent, inspects the complete diff, prioritizes concrete defects, verifies tests, and leaves merge authority with the team.
Read guideProduct workflowAgent skills for product managers
Turn recurring product work into reviewable Agent Skills with explicit evidence, decision rights, templates, uncertainty handling, and human approval around roadmap commitments.
Read guide