Integrate with an AI coding agent

Choose one documentation source for your agent. The Legnext Skill is the recommended path; the docs MCP server is the live alternative. You do not need both.

1. Choose a documentation source

Legnext Skill — recommended

Install the integration playbook in Claude Code, Cursor, Codex, and other skill-compatible agents.

Docs MCP server

Give any MCP-capable client live documentation search without installing a project skill.

Use the Legnext Skill

Run this from your project:

npx skills add legnext-ai/skills

The skill teaches the task lifecycle, retries, output persistence, child tasks, credits, and error handling.

Use the docs MCP server

Add this URL as a Streamable HTTP MCP server named legnext-docs:

https://docs.legnext.ai/mcp

For Claude Code:

claude mcp add --transport http legnext-docs https://docs.legnext.ai/mcp

The docs server does not require a Legnext API key. If your agent supports neither Skills nor MCP, tell it to fetch llms-full.txt.

2. Add your API key

Set LEGNEXT_API_KEY in the server-side environment of your project. Never paste the key into the agent conversation, browser code, or a committed file.

3. Paste this prompt

Integrate the Legnext API into this project.

Inspect the repository and its local instructions before changing code. Preserve
the existing architecture. Use LEGNEXT_API_KEY only from the server environment
and keep it behind a server boundary.

Add an image-generation flow that submits a diffusion task, immediately
persists its job_id, waits for a terminal state, downloads the result before its
URL expires, and returns or displays the saved image. Pin Midjourney version 8.2
and use a 16:9 aspect ratio.

Use the installed Legnext skill or the legnext-docs MCP server as the source of
truth. Do not invent request fields. Handle rejected submits, failed tasks,
the retry status, client timeouts, ambiguous submit failures, duplicate attempts,
and output persistence. Keep the Get Task capability URL server-side. Add and
run a focused verification, then report the changed files and how to try the
integration.

4. Verify the result

The finished integration should:

Keep the API key at a server boundary
Persist the first returned job_id instead of resubmitting uncertain requests
Continue through all non-terminal states and stop at completed or failed
Keep the unauthenticated Get Task capability URL private
Copy completed output to durable storage
Handle moderation, balance, timeout, and provider failures
Include a focused test or repeatable verification

Prefer to write the integration yourself? Follow First API request, then read Task lifecycle.

Start here

API basics

Reference

Platform integrations

API reference

Updates

Integrate with an AI coding agent

1. Choose a documentation source

Legnext Skill — recommended

Docs MCP server

Use the Legnext Skill

Use the docs MCP server

2. Add your API key

3. Paste this prompt

4. Verify the result

​1. Choose a documentation source

Legnext Skill — recommended

Docs MCP server

​Use the Legnext Skill

​Use the docs MCP server

​2. Add your API key

​3. Paste this prompt

​4. Verify the result

1. Choose a documentation source

Use the Legnext Skill

Use the docs MCP server

2. Add your API key

3. Paste this prompt

4. Verify the result