Turn 10,000 API endpoints into one CLI tool instead of MCP, Skills and tools zoo

As AI agents proliferate, developers face a tooling crisis: wiring up hundreds of MCP servers, skills, and agent tools creates unsustainable complexity. The new openapi-to-cli tool offers a radical alternative by converting OpenAPI/Swagger specifications into unified command-line interfaces at runtime. Instead of exposing thousands of individual tools that consume valuable context window space with JSON schemas, this approach provides a single shell execution tool. Agents can discover commands through natural language search using a BM25 engine (ported from picoclaw) that ranks across command names, paths, descriptions, and parameters.

The system caches API specs locally and supports multiple profiles per API—like read-only versus admin access—while allowing completely different APIs to be mounted into the same binary. This means an agent can switch between GitHub and Box APIs with simple commands like 'ocli use github' followed by 'ocli commands --query "create pull request".' The tool generates commands dynamically from cached specifications, eliminating code generation and compilation overhead. For large APIs with hundreds of endpoints, traditional --help becomes useless, but the built-in search functionality lets agents find relevant operations efficiently.

This CLI-wrapper pattern represents a significant architectural shift for AI agent tooling. By reducing tool metadata overhead and providing intelligent discovery mechanisms, it enables agents to work with massive APIs like GitHub's REST interface (with hundreds of endpoints) without drowning in schema definitions. The approach keeps agent context focused on reasoning rather than tool descriptions, potentially improving performance and reliability in complex automation scenarios.