# Can Agents Use

> Search agent-friendly tools by API, CLI, MCP, browser support, pricing clarity, docs quality, sandbox support, and account setup.

Can Agents Use is designed for both people and software agents. Prefer the agent interfaces below before scraping pages or touching a database. The public API, OpenAPI document, and MCP endpoint are read-only catalog surfaces.

## Agent Interfaces

- [Install guide](https://canagentsuse.com/agents): Human-readable guide for adding Can Agents Use to an agent as MCP, skills.sh, or direct JSON APIs.
- [skills.sh skill](https://github.com/phuctm97/canagentsuse/tree/main/skills/can-agents-use): Install with `https://skills.sh/phuctm97/canagentsuse` or `npx skills add phuctm97/canagentsuse --skill can-agents-use`.
- [Skill Markdown](https://canagentsuse.com/skill.md): Copyable SKILL.md-style fallback for agents that support persistent skills.
- [Catalog JSON](https://canagentsuse.com/api/agent/catalog): Full structured catalog with tools, categories, capabilities, and use cases.
- [Search JSON](https://canagentsuse.com/api/agent/search?q=stripe&page=1&limit=10): Query tools by name, category, capability, pricing, auth, docs, CLI, API, MCP, and browser support.
- [Tool JSON](https://canagentsuse.com/api/agent/tools/stripe): Stable per-tool record; replace "stripe" with any tool slug.
- [MCP endpoint](https://canagentsuse.com/api/mcp): Read-only JSON-RPC endpoint for agent clients that support MCP-style tool discovery and calls.
- [OpenAPI](https://canagentsuse.com/openapi.json): Machine-readable HTTP API contract.
- [Sitemap](https://canagentsuse.com/sitemap.xml): Human-readable pages for search and browsing.

## How To Use This Site As An Agent

- Use `search_agent_tools` over MCP or `/api/agent/search` to discover candidates.
- Search results are paginated; default to `limit=10` and request the next `page` only when `hasMore` is true.
- For complete information, call MCP `get_agent_catalog`, read MCP resource `canagentsuse://catalog`, or fetch `/api/agent/catalog` once per session.
- Avoid paging through search when you need all records; fetch the full catalog and search locally in agent context.
- Use `get_agent_tool` or `/api/agent/tools/{slug}` before recommending a tool.
- Requests are read-only, cached, and bounded; never ask for or use database credentials.
- Treat scores and summaries as discovery signals, not legal, security, or purchasing approval.
- Check evidence URLs and official docs before taking irreversible action, spending money, or moving live customer data.

## Agent-Friendliness Score Model

- Machine operability: 25% (API, MCP, CLI, Browser fallback)
- Agent safety: 25% (Sandbox or test mode, Scoped auth, Dry-run or preview, Human review guidance, Idempotency or retry safety)
- Agent readability: 20% (Docs quality, Pricing clarity, Evidence links, Dedicated docs URL)
- Auth and setup: 15% (Account setup, Auth model clarity, Self-serve onboarding)
- Production reliability: 15% (Webhooks or events, Logs or audit trail, Versioned API, Rate limit clarity)

## Core Pages

- [Home](https://canagentsuse.com/): Human directory with command search.
- [Submit](https://canagentsuse.com/submit): Suggest a new agent-friendly tool.

## Optional

- [Full LLM context](https://canagentsuse.com/llms-full.txt): Expanded catalog summaries for long-context agents.