Coding agent support and tools

Overview

LiveKit is built for coding agents like Claude Code , Cursor , Codex , and Gemini CLI . Use the resources in this guide to give them direct access to the latest docs and best practices so they build the right thing every time.

For optimal performance, your coding agent needs three types of resources:

1. Best practices in-context — With Agent Skills and AGENTS.md files, your coding agent knows how to approach each part of building an agent, and always consults the latest documentation when building with LiveKit.
2. Docs and code access — With the LiveKit CLI or Docs MCP server, your coding agent can browse and search the LiveKit documentation and SDK code to find up-to-date information and advice.
3. A way to try its changes — With the agent debugger, your coding agent can run your agent locally and hold a conversation with it, so it checks what a change does instead of guessing.

This guide covers each of these areas in detail. The LiveKit starter projects come with these resources built-in, including a robust AGENTS.md file:

Python starter project

Includes an AGENTS.md file optimized for building agents in Python.

livekit-examples/agent-starter-python

Node.js starter project

Includes an AGENTS.md file optimized for building agents in Node.js.

livekit-examples/agent-starter-node

Agent Skills

LiveKit publishes Agent Skills  that teach coding agents how to build, test, and ship voice agents with the LiveKit Agents SDK. The skills cover the whole development loop, from architecture and live debugging to tests, simulations, and deployment. Your coding agent picks up the right skill from what you ask for, and each skill sends it to the docs for API details, so skills work best alongside docs search.

Install the skills for the coding agents on your machine with the skills CLI :

npx skills add livekit/agent-skills

The skills are also available as a Claude Code plugin, which sets up the Docs MCP server too. For every install option, the full list of skills, and how updates work, see the Agent Skills reference.

Docs search

Docs search is available in both the LiveKit CLI and the LiveKit Docs MCP server. Both have the exact same capabilities — choose the one that works best for your agent. All documentation files are also available in raw Markdown and via LLMs.txt.

LiveKit CLI

The LiveKit CLI includes a docs subcommand that gives any coding agent full access to LiveKit documentation from the terminal. It works with any agent that can execute shell commands — browse the table of contents, search docs, read full pages, and search code across LiveKit repositories.

Add the following to your AGENTS.md to ensure your agent always consults the docs:

LiveKit is a fast-evolving project. Always refer to the latest documentation. Run `lk docs --help` to see available commands. Key commands: `lk docs overview`, `lk docs search`, `lk docs get-page`, `lk docs code-search`, `lk docs changelog`, `lk docs pricing-info`. Run `lk docs <command> --help` before using a command for the first time. Prefer browsing (`overview`, `get-page`) over search, and `search` over `code-search`, as docs pages provide better context than raw code.

MCP server

For coding agents that support Model Context Protocol (MCP) , LiveKit provides a free MCP server with tools for browsing and searching the docs site. The server supports the same capabilities as the CLI with deeper IDE integration.

Cursor

Claude Code

VS Code

Codex

CursorClaude CodeVS CodeCodex

Add this to your MCP configuration:

{"livekit-docs": {"url": "https://docs.livekit.io/mcp"}}

Or click this button:

claude mcp add --transport http livekit-docs https://docs.livekit.io/mcp

The LiveKit plugin for Claude Code includes the MCP server, so you can skip this step if you install it.

code --add-mcp '{"name":"livekit-docs","type":"http","url":"https://docs.livekit.io/mcp"}'

codex mcp add --url https://docs.livekit.io/mcp livekit-docs

Add the following to your AGENTS.md to ensure your agent always consults the docs:

LiveKit is a fast-evolving project. Always refer to the latest documentation. LiveKit provides an MCP server at `https://docs.livekit.io/mcp` with tools for browsing and searching docs. Key tools: `get_docs_overview`, `get_pages`, `docs_search`, `code_search`, `get_changelog`, `get_pricing_info`. Prefer browsing (`get_docs_overview`, `get_pages`) over search, and `docs_search` over `code_search`, as docs pages provide better context than raw code.

Markdown docs and LLMs.txt

Each page on the LiveKit docs site is available in Markdown format, which is useful for pasting into AI assistants when MCP and the CLI aren't available.

To access the Markdown version of any page, append .md to the end of the URL or add text/markdown to your Accept header. For example, this page is available at https://docs.livekit.io/intro/coding-agents.md . You can also use the "Copy page" button on the top right of any docs page.

An index of the docs site is available at https://docs.livekit.io/llms.txt . It links to a separate index for each documentation section, and each section index lists every page in that section with a brief description. To get a section index directly, append /llms.txt to one of those section URLs, for example https://docs.livekit.io/telephony/llms.txt .

Every page of the site in a single file is available at https://docs.livekit.io/llms-full.txt , but this file is very large and is not suitable for all use cases. For more information about LLMs.txt, see llmstxt.org .

Agent debugger

The LiveKit CLI includes lk agent debugger, a text-mode harness that lets a coding agent live-test the agent it's building. The debugger runs your agent locally in the background, and the coding agent plays the user one turn at a time. Each turn prints the agent's reply along with the tool calls, handoffs, and errors that produced it, so the coding agent can see why the agent did what it did, not only what it said.

The debugger doesn't use audio or a LiveKit room, so a turn costs only your agent's own LLM and tool calls. That makes it cost-effective enough to run after every change.

lk agent debugger start
lk agent debugger say "Hi, what can you do?"
lk agent debugger say "Book me a table for two tonight"
lk agent debugger restart   # after editing the agent's code
lk agent debugger stop

The debugging-livekit-agents skill teaches your coding agent to use the debugger, and it's included when you install the skills. Or add the following to your AGENTS.md so your coding agent tries its changes:

After changing the agent, try it with `lk agent debugger` before calling the change done. Start the agent with `lk agent debugger start`, send user turns with `lk agent debugger say "..."`, and read the tool calls in each turn as well as the reply. Run `lk agent debugger restart` after every code edit, since a running session keeps the old code. Use `say --logs` to see a tool's traceback next to the turn that raised it, and `lk agent debugger stop` when done.

Best practices

To construct your own AGENTS.md file or agent skills, consider the following best practices to help your coding agent get the most out of LiveKit's tools:

• Begin with a starter project. The Python  and Node.js  starters include working agents, tests, and an optimized AGENTS.md.
• Read the docs like a human: browse the table of contents first (lk docs overview), search docs second (lk docs search), and search code third (lk docs code-search). Browsing gives full context — search only gives fragments.
• Always check the docs before writing LiveKit code. The APIs change frequently and training data goes stale.
• Try every change before calling it done. Drive a few turns with the agent debugger and read the tool calls, not only the reply. Restart the debugger after each edit.
• Use code search to answer detailed questions about a class or method that isn't present in the docs.
• If the docs don't match the package installed or something breaks after an upgrade, check the changelog (lk docs changelog).
• Search results only show excerpts — always fetch the full page with lk docs get-page to see prerequisites and related options.
• Practice TDD with the agent testing tools in the Python and Node.js Agents SDKs. The testing guide also has advice on CI setup.