Skills for coding agents

Collimate is built for agents to drive. If you are wiring up an AI coding agent — Claude Code, Cursor, or your own — start here.

The fastest path: the hosted MCP endpoint

Add the hosted MCP endpoint and your agent gets sandbox_create, sandbox_exec, sandbox_write_file, sandbox_fork, sandbox_suspend, sandbox_resume, sandbox_list, and sandbox_delete as native tools — no SDK, no glue code.

claude mcp add --transport http collimate https://mcp.collimate.ai \
  --header "Authorization: Bearer $COLLIMATE_API_KEY"

Drop in the Collimate skill

For agents that use skill files, drop SKILL.md into your agent's skills directory. It teaches the create → exec → fork → suspend workflow, the REST surface, the MCP tools, and the one idea worth internalizing — prepare once, fork many — so the agent reaches for live fork instead of rebuilding state per branch.

Why fork is the agent move

The highest-leverage thing an agent can do on Collimate is fork a prepared sandbox to try several actions in parallel and keep the best — instead of booting a fresh environment and replaying its history for every branch. The skill file makes that the default. See Live fork.

Agent-readable docs

• /llms.txt — a curated map of this site for LLMs, with the skill file first. Point your agent's retrieval at it.
• /openapi.yaml — the machine-readable API description, for agents that generate clients or tool schemas.
• Every page here is written to be read by an agent: honest field tables, real request/response shapes, and stable error codes to branch on.

Markdown source of any page

Prefer machine-readable docs? The llms.txt + SKILL.md pair is the supported agent-retrieval surface. Fetch the curated llms.txt map above; it points an agent at everything it needs.

Start here

MCP quickstart Add the endpoint to Claude Code or Cursor.

The SKILL.md file Drop-in skill teaching an agent the API + MCP.

Live fork The prepare-once, fork-many primitive.

API reference Every endpoint and field.