MCP & Skills
Model Context Protocol (MCP) is the open standard for connecting language models to external tools and data. Skills are scoped knowledge packs the model can invoke on demand. Ptah uses both to deliver a consistent, high-leverage experience across every provider — Claude, Gemini, Codex, Copilot, and local CLI agents alike.
The two pillars
Section titled “The two pillars”MCP — tools the model can call
Section titled “MCP — tools the model can call”Ptah ships with a built-in MCP server exposing the ptah_* tool family: workspace analysis, diagnostics, file search, browser automation, git worktree management, agent spawning, and more. You can also plug in third-party MCP servers (filesystem, databases, SaaS APIs) via configuration.
Skills — knowledge the model can invoke
Section titled “Skills — knowledge the model can invoke”Skills are versioned, file-based prompt packages. Each has a SKILL.md that tells the orchestrator when to invoke it and what context it adds. Ptah auto-publishes plugin skills as symlinks under <workspace>/.claude/skills/ so third-party AI clients (Copilot, Cursor, Codex CLI) can discover the same knowledge.
Why this matters
Section titled “Why this matters”Without MCP and Skills, a model depends on whatever its provider gave it — Claude knows nothing about Copilot’s tools, and vice versa. Ptah solves this with a unified layer:
- One tool catalog across all providers via the built-in MCP server.
- One skill catalog across all AI clients via
.claude/skills/junctions. - One extension mechanism — plugins contribute both.
Explore the docs
Section titled “Explore the docs”| Topic | Link |
|---|---|
| The built-in MCP server and Code Execution | Built-in MCP server |
Full catalog of ptah_* tools | Ptah tools |
| What skills are and how junctions work | Skills |
| Pre-curated skill catalog | Popular skills |
| Authoring your own skills | Creating skills |
| Connecting third-party MCP servers | Third-party MCP |