This page helps you set up your coding agent with the tools and context it needs to work with Extend.
Extend ships purpose-built artifacts for each style. Match the one(s) your agent will use, then install only those.
Most teams set up all three. They’re additive, and steps 3–5 below cover each.
The agents and harnesses this works with include Claude Code, Cursor, Codex, OpenCode, and Goose — anything that can read a skill file or fetch a URL.
Your agent’s calls — CLI or SDK — fail without a key. Create one from the Developers dashboard and export it where your agent’s shell can see it:
The CLI reads EXTEND_API_KEY automatically. With an SDK, pass the key when you create the client (reading it from the environment is common) — the SDK then sets the auth header and API version for you. If your agent makes raw HTTP requests instead, it must send both an Authorization: Bearer <key> header and x-extend-api-version: 2026-02-09 on every request.
Install the CLI:
Confirm it’s there with extend --version. The CLI covers parse, extract, classify, split, edit, run, and batch, plus management of the resources behind them (runs, files, extractors/classifiers/splitters/workflows, webhooks).
Then install the agent skill, which is generated from the CLI’s own command tree and follows the agentskills.io standard. It teaches the harness how to drive extend correctly:
Writes ~/.agents/skills/extend/SKILL.md, the cross-client default path.
The skill is generated from the installed CLI, so re-run extend skill install after upgrading the CLI to keep it in sync.
When your agent generates application code, drop Extend’s platform context file at the repo root. It’s the same content as the AGENTS.md / CLAUDE.md convention — a compact brief on the API, SDKs, schema rules, sync-vs-async patterns, webhooks, and error handling — so the SDK code your agent writes is correct on the first try:
Prefer it available across projects? Save it as a reusable skill under ~/.agents/skills/ instead of per-repo.
The official SDKs cover four languages. Python, TypeScript, and Java ship polling helpers (create_and_poll / createAndPoll) and built-in webhook signature verification; Go has neither, so an agent writing Go should poll and verify signatures manually.
Whichever path you chose, your agent can pull authoritative reference on its own:
.md to a doc URL (e.g. https://docs.extend.ai/extraction/schema.md) for clean, token-cheap content.Add these to your agent’s allowed-fetch list or mention them in your project instructions so it reaches for the docs instead of guessing.