integrations/official/AIgateway Skill

AIgateway Skill + AIgateway

Install AIgateway as a skill your coding agent loads — the discovery loop, baked in.

A retrieval-biased agent skill: instead of pasting docs into context, your agent loads one SKILL.md that teaches it to discover models live (get_model / /v1/models/{id}/schema) and call any modality across 1000+ models. It hardcodes no catalog, so it never goes stale.

AIgateway Skill homepage →

Setup

Three steps or fewer.

STEP 01

Any harness — drop in the skill file

Save the canonical skill into your harness's skills directory. For Claude Code that's ~/.claude/skills/aigateway/SKILL.md. The file is regenerated from the live catalog, so re-fetch it any time.

mkdir -p ~/.claude/skills/aigateway
curl -fsSL https://aigateway.sh/skill.md -o ~/.claude/skills/aigateway/SKILL.md

STEP 02

Claude Code — install the plugin (skill + MCP in one step)

The plugin bundles the skill and auto-wires the MCP server, so the agent gets the instructions and the live tools (list_models, get_model, chat, …) together.

/plugin marketplace add aigateway-sh/mcp-server
/plugin install aigateway

STEP 03

Verify the agent discovers, not guesses

Ask it to call list_models for a modality, then get_model on a slug. It should read the exact request/response schema before making a real call.

Notes

Published in the machine-readable index at /.well-known/agent-skills/index.json (with a content digest) so agents can auto-discover and verify it.
The skill teaches the discovery loop; for the full per-endpoint reference it points the agent at /reference.md and /agents.md.

AIgateway Skill + AIgateway

Three steps or fewer.

Any harness — drop in the skill file

Claude Code — install the plugin (skill + MCP in one step)

Verify the agent discovers, not guesses

Same key. Every other tool.