llms.txt - For AI Agents & Answer Engines

Swytchcode publishes a canonical /llms.txt file at the root of this domain for AI agents, answer engines, and crawler tools.

It is a plain-text, machine-readable summary of Swytchcode’s current developer experience, including:

• CLI installation and authentication workflows
• Managed Authentication and credential resolution
• Runtime tool execution pipeline and execution policies
• Retries, timeouts, and idempotency for safety
• Runtime SDKs for JavaScript and Python
• MCP server setup and API access for agents
• Open source runtime repos and integration docs

Why llms.txt matters

AI systems increasingly answer directly from structured documentation instead of just ranking links. /llms.txt serves as an authoritative summary layer that tells agents:

• what Swytchcode does
• how to install and run the CLI
• how to connect authenticated providers
• how tools execute reliably through managed runtime behavior
• where the canonical docs and source code live

This helps AI assistants return accurate answers and routes them to the right docs for Swytchcode.

What this page documents

This page collects the latest Swytchcode documentation topics relevant to AI agents and trusted tool execution.

• CLI Overview - Install the Swytchcode CLI, authenticate, and start executing tools.
• Authentication - Learn how account and provider authentication works.
• Managed Authentication - Secure credential storage and provider credential resolution.
• Execution Pipeline - The full runtime flow from tool resolution to API response.
• Retries - How Swytchcode retries transient failures automatically.
• Idempotency - Prevent duplicate operations during retries.
• Workspaces and teams - Personal and team workspaces, roles, seats, and per-workspace log storage.
• Pricing - Current plans, limits, and billing on the Swytchcode website.
• Policies Overview - Policy guardrails that validate every tool execution.
• Policy recipes - Ready-to-use policies for email, payments, data safety, code, time windows, and more.
• Human approval - Hold high-impact commands until someone approves or denies them in Slack.
• tooling.json Reference - Define trusted methods.
• manifest.json Reference - Configure endpoints, auth, retries, and execution policies.
• policies.json Reference - Apply runtime guardrails and validation rules.
• Runtime SDK Overview - JavaScript and Python SDKs for tool execution.
• JavaScript SDK - Build Node.js/Bun/Next.js agents with the runtime.
• Python SDK - Execute trusted tools from Python applications.
• Anthropic SDK Quickstart - Connect Claude agents to trusted tools.
• Native SDK Quickstart - Use Swytchcode tools with provider SDKs and platforms.
• MCP Server Reference - Set up and connect the MCP server for Cursor, Claude Code, Windsurf, Copilot, Gemini CLI, Codex, Hermes, and OpenClaw.
• MCP Tool Reference - The full tool-by-tool table for both MCP profiles (agent: 7 tools, full: 15 tools) and their CLI equivalents.
• CLI Command Reference - Every swy subcommand, flag, and exit code.
• Troubleshooting - Exit codes, common failure modes, and diagnostic commands.
• Cookbook - Real-world agent example projects with source on GitHub.

What /llms.txt includes

The published file includes:

• Product description and canonical doc links
• CLI commands for install, login, auth, and mcp management
• MCP tool profiles (agent vs full) and their per-tool CLI equivalents
• Guidance for creating and using trusted tools
• Notes on managed auth, retries, idempotency, and policies
• Links to runtime SDK docs and provider integration guides
• Community and open source links for Swytchcode

Connecting agents via MCP

The MCP server exposes two tool profiles, selected with --profile:

swy mcp serve --profile agent   # default: 7 tools, execution-focused (discover, info, exec, list, search, add, policy-read)
swy mcp serve --profile full    # 15 tools: adds init, bootstrap, version, get, doctor, policy_check, policy_add, policy_remove

Most MCP-compatible editors (Cursor, Windsurf, Copilot, Gemini CLI, Codex, Hermes, OpenClaw) register the server automatically via swy init --editor=<name>. Claude Code is the exception - it’s registered separately with claude mcp add. Once connected, Swytchcode handles input validation, credential resolution, policy evaluation, retries, idempotency, and response normalization automatically for every tool call.

See the MCP server reference for setup steps, the MCP tool reference for the full tool table, and Troubleshooting if a connection or execution fails.

Browse the full catalog of supported APIs at swytchcode.com/apis.

Open source and GitHub

Swytchcode runtime components are open source and maintained in GitHub repositories, including runtime SDKs for JavaScript and Python.

Direct link

https://docs.swytchcode.com/llms.txt