Getting Started

Summary

A quick assessment and prioritized roadmap for making any codebase agent-ready in weeks, not months. Evaluate seven key questions about discovery, APIs, CLI output, authentication, error handling, and testing. Then implement in priority order, starting with low-effort high-impact items like AGENTS.md and --json output.

• Discovery (llms.txt, AGENTS.md)
• CLI JSON output and machine-readable formats
• Structured error handling (RFC 7807)
• OAuth 2.1 Client Credentials authentication
• MCP server exposure
• Workflow definitions (Arazzo)
• Evaluation suite (evals, metrics, red-teaming)

Quick Assessment

Before diving into specific dimensions, assess where your codebase stands:

1. Can an AI agent discover your capabilities? (Discovery & AEO)
2. Can an AI agent call your APIs programmatically? (API Surface)
3. Does your CLI produce machine-readable output? (CLI Design)
4. Do you expose an MCP server? (MCP Servers)
5. Can agents authenticate without browser flows? (Authentication)
6. Do your errors tell agents how to recover? (Error Handling)
7. Can you verify agent interactions work? (Testing)

Priority Order

If you're starting from zero, implement in this order:

1. Write an AGENTS.md and llms.txt file
2. Add --json output to your CLI
3. Implement RFC 7807 structured errors
4. Switch to OAuth 2.1 Client Credentials for M2M auth
5. Expose an MCP server for your API
6. Add Arazzo workflow definitions
7. Build an evaluation suite

Each subsequent section of this reference covers one dimension in depth.