Use with your coding agent¶

pontifex-mcp bundles an official Agent Skill in the package: a SKILL.md that teaches a coding agent how to build with it. The skill covers the PontifexMCP facade, namespace:resource:action scopes, API-key and JWT auth, the pontifex-mcp CLI, caching, and generating tools from an OpenAPI spec.

The skill ships and versions with the library, so your agent reads guidance that matches the version you installed. Upgrade the package and the guidance moves with it.

Install the skill¶

Add the package, then run library-skills to wire the skill into your project:

uv add pontifex-mcp        # or: pip install pontifex-mcp
uvx library-skills         # scans your dependencies, installs the skills you pick

or just install the specific skill for pontifex mcp

uvx library-skills -s pontifex-mcp --yes

library-skills scans your installed dependencies, finds the ones that ship a skill, and links each into .agents/skills/, the shared directory that Codex, Cursor, GitHub Copilot, OpenCode, and other agents read. That one command covers them all.

Claude Code reads its own .claude/skills/ directory, so it takes a dedicated flag --claude:

uvx library-skills -s pontifex-mcp --claude --yes

Either way, the link points back at the installed package, so the skill stays current as you upgrade.

What your agent gets¶

The skill loads when you ask your agent to work with pontifex-mcp. It walks the build path: PontifexMCP, @mcp.tool(scope=...), ApiKeyAuth and JwtAuth, the CLI, caching, and OpenAPI tool generation. It also pins down the conventions an agent tends to miss from memory, like the decorator needing parentheses and the SQLite-to-Postgres switch being a config change. For moving across breaking releases, it carries the upgrade steps.