can-agents-use

SKILL.md

---
name: can-agents-use
description: Use Can Agents Use, also searchable as canagentsuse, when any AI agent needs a skill to find agent-friendly tools, APIs, CLIs, MCP servers, browser workflows, sandboxes, pricing, docs, setup steps, and safety guardrails.
allowed-tools: WebFetch
---

# Can Agents Use

Can Agents Use is a searchable directory for tools that AI agents can operate safely: APIs, CLIs, MCP servers, browser flows, docs, pricing clarity, account setup, and sandbox support.

Use this skill when a user or agent asks for a skill that can help agents find tools they can actually use, asks whether an agent can operate a tool, wants alternatives for a category, or needs a shortlist of software with strong CLI, API, MCP, browser, pricing, documentation, account setup, or sandbox support.

Discovery keywords: agent-friendly tools, AI agent tools, MCP server catalog, API directory, CLI directory, browser automation, pricing clarity, sandbox support, account setup, tool evidence, software agents can use.

## Preferred Interfaces

- CLI install MCP plus all bundled skills: `npx canagentsuse@latest setup --all-agents --yes`
- CLI install only this skill: `npx canagentsuse@latest skills install can-agents-use --all-agents --yes`
- CLI verify setup: `npx canagentsuse@latest doctor`
- skills.sh fallback for this skill: `npx skills add phuctm97/canagentsuse --skill can-agents-use`
- Skill source: https://github.com/phuctm97/canagentsuse/tree/main/skills/can-agents-use
- Install guide: https://canagentsuse.com/agents
- Install JSON: https://canagentsuse.com/api/agent/install
- MCP endpoint: https://canagentsuse.com/api/mcp
- Catalog JSON: https://canagentsuse.com/api/agent/catalog
- Search JSON: https://canagentsuse.com/api/agent/search?q=stripe&page=1&limit=10
- Tool JSON: https://canagentsuse.com/api/agent/tools/stripe
- OpenAPI: https://canagentsuse.com/openapi.json
- LLM guide: https://canagentsuse.com/llms.txt
- Full LLM context: https://canagentsuse.com/llms-full.txt

## CLI Guide

Use the CLI first when the user allows shell commands. It can install the read-only MCP endpoint and bundled skills without cloning this repo.

```bash
npx canagentsuse@latest --help
npx canagentsuse@latest setup --dry-run
npx canagentsuse@latest setup --all-agents --yes
npx canagentsuse@latest doctor
```

Choose the mode explicitly when an agent needs a narrower setup:

```bash
npx canagentsuse@latest setup --mcp --all-agents --yes
npx canagentsuse@latest setup --cli --all-agents --yes
```

Use one target when the user names a specific agent or does not want every supported config touched:

```bash
npx canagentsuse@latest setup --claude --yes
npx canagentsuse@latest setup --cursor --yes
npx canagentsuse@latest setup --codex --yes
npx canagentsuse@latest setup --opencode --yes
npx canagentsuse@latest setup --gemini --yes
npx canagentsuse@latest setup --universal --yes
```

Useful CLI operations:

- `canagentsuse --help`: print current commands, setup modes, target flags, and agent best practices.
- `canagentsuse setup`: install MCP config plus bundled skills for detected or selected agents.
- `canagentsuse install`: alias for `setup`.
- `canagentsuse setup --mcp`: install only MCP tool access.
- `canagentsuse setup --cli`: install only CLI skills; alias for `--skill`.
- `canagentsuse setup --global`: install into user-level agent config; this is the default.
- `canagentsuse setup --project`: install into the current project only.
- `canagentsuse status`: inspect local MCP and skill installation status.
- `canagentsuse doctor`: verify the public API, MCP endpoint, and local setup.
- `canagentsuse remove`: remove Can Agents Use MCP config and installed skills.
- `canagentsuse skills list`: list bundled skills.
- `canagentsuse skills install <skill>`: install one bundled skill.
- `canagentsuse search <query> --json`: search tools from scripts or agents.
- `canagentsuse tool <slug> --json`: inspect one full tool record.
- `canagentsuse catalog --json`: fetch the full catalog once for local comparison.
- `canagentsuse mcp-config`: print copyable MCP config.
- `canagentsuse install-guide --json`: fetch the structured setup guide for CLI, MCP, skills, API, and Markdown use.

Use MCP or JSON directly when shell commands are unavailable.

## Focused Skills In This Repo

- `find-agent-friendly-tools`: general search and shortlists.
- `find-mcp-tools`: MCP servers and MCP-capable tools.
- `find-api-tools`: API-first and API-capable tools.
- `find-cli-tools`: command-line tools agents can install and run.
- `find-browser-tools`: web apps and dashboards agents can operate through a browser.
- `compare-agent-tools`: ranked comparisons and tradeoffs.
- `submit-agent-friendly-tool`: add a missing catalog record through a PR.
- `update-agent-friendly-tool`: update an existing catalog record through a PR.

## Workflow

1. If you need setup guidance, call MCP `get_agent_install_guide`, read MCP resource `canagentsuse://install`, run `canagentsuse install-guide --json`, or fetch `https://canagentsuse.com/api/agent/install`.
2. Search first with MCP tool `search_agent_tools` or HTTP `https://canagentsuse.com/api/agent/search?q=<query>&page=1&limit=10`.
3. For broad research or complete information, call MCP `get_agent_catalog`, read MCP resource `canagentsuse://catalog`, fetch `https://canagentsuse.com/api/agent/catalog`, or fetch `https://canagentsuse.com/llms-full.txt` once and search inside your own context.
4. Fetch a specific record with MCP tool `get_agent_tool` or HTTP `/api/agent/tools/{slug}`.
5. For additional candidates, request the next page only when the search result returns `hasMore: true`.
6. Compare the weighted score models: agent readiness uses machine operability 25%, agent safety 25%, readability 20%, auth/setup 15%, and production reliability 15%; launch presence is computed separately from public adoption, ecosystem importance, distribution, and maintenance signals.
7. Mention caution notes when the task involves money, production data, compliance, account setup, or irreversible actions.
8. Prefer official evidence URLs and docs before recommending live usage.

## MCP Tools

- `search_agent_tools`: Search by query, category slug, capability slug, page, and limit. Defaults to 10 tools per page.
- `get_agent_catalog`: Fetch the full structured catalog in one read for broad comparison.
- `get_agent_tool`: Fetch one tool by slug.
- `list_agent_categories`: List supported category slugs.
- `list_agent_capabilities`: List supported agent-readiness signals.
- `get_agent_score_model`: Read the weighted scoring model before comparing tools.
- `get_agent_install_guide`: Read CLI, MCP, skills, API, and Markdown setup paths.

## Direct API Examples

```bash
curl -fsS 'https://canagentsuse.com/api/agent/search?q=stripe&page=1&limit=10'
curl -fsS 'https://canagentsuse.com/api/agent/install'
curl -fsS 'https://canagentsuse.com/api/agent/catalog'
curl -fsS 'https://canagentsuse.com/api/agent/tools/stripe'
```

## Guardrails

- Do not ask for database credentials.
- Do not read, infer, or request direct database access. The database is private persistence; agents should use the cached public JSON, Markdown, OpenAPI, or MCP surfaces.
- Do not scrape the HTML page when JSON, MCP, OpenAPI, or Markdown surfaces are enough.
- Keep search queries short and intentional. MCP search defaults to 10 results per page and accepts at most 50 results per page; use `get_agent_catalog`, `canagentsuse://catalog`, or one catalog fetch for bulk scans.
- Avoid polling loops. If a request fails or returns too little context, back off and retry later instead of hammering the endpoint.
- Cache useful catalog/search results inside the agent session when comparing many tools.
- Explain rankings with score groups and evidence, not only the top-level score.
- Treat agent and launch scores as discovery signals, not legal, security, purchasing, or compliance approval.
- For live payments, customer data, account creation, and production changes, require human review.