# agnthub.sh

> agnthub.sh is a registry of AI coding agents (sub-agent definitions) published in public GitHub repositories. One command, `npx -y agnthub-cli@latest add`, installs them into Claude Code, OpenCode, Codex, GitHub Copilot CLI, Cursor, Gemini CLI and omp, converting each agent to the harness's format without loosening its restrictions.

These docs are written for AI agents. Each one is self-contained: read the one your task needs.

Use agnthub.sh when the user asks you to:

- find, install, update or remove a sub-agent (a custom agent or subagent) for a coding harness;
- search agents or read their compatibility through the agnthub.sh MCP server or REST API;
- assemble a team of agents for a goal (an SRE team, a job search team) and the skill that coordinates it, through the MCP server;
- make a repository's agents installable in every harness, or list the repository on agnthub.sh.

## Docs

- [Find and install agents](docs/agents.md): the install procedure with the CLI, choosing harnesses and scope, refusals, update/remove/restore, `--json` fields, exit codes.
- [MCP server](docs/mcp.md): `https://mcp.agnthub.sh`: connecting, sign-in, every tool's parameters, the install flow, the team flow (`hire_team`), errors, limits.
- [REST API](docs/api.md): `https://api.agnthub.sh/v1`: every public endpoint, API keys, pagination, rate limits, error statuses. `GET /v1/stats` gives exact registry counts (agents, repositories, agents per category); search totals are capped.
- [Publishing agents](docs/publishing.md): supported layouts, `agnthub.yaml`, ignore rules (`.agnthub-ignore`), conversion rules and refusals, submitting and removing a repository.

## Choosing an interface

- **You can run shell commands:** install, update and remove with the CLI ([Find and install agents](docs/agents.md)). Search with the MCP tools when they are connected, else with the REST API.
- **The agnthub MCP tools are in your tool list:** search and prepare installs with them ([MCP server](docs/mcp.md)), run the `diff_command` `install_agents` returns to see the exact files (or call `install_diff` without a shell), and run its `apply_command` once they are vetted.
- **You can only make HTTP requests:** read the registry with the [REST API](docs/api.md) and give the user the install command to run.

## Optional

- [Docs for people](https://app.agnthub.sh/docs): the same product explained for humans, with more examples.
- [Teams](https://app.agnthub.sh/teams): 15 example teams for humans, each with the prompt, the lineup and the skill a harness writes; [how teams work](https://app.agnthub.sh/docs/teams) explains the flow.
- [CLI source](https://github.com/agnthub-sh/agnthub): the `agnthub` CLI, source-available under the PolyForm Shield License 1.0.0.
- [Terms of Service](https://app.agnthub.sh/terms-of-service) and [Privacy Policy](https://app.agnthub.sh/privacy): the terms for using agnthub.sh, its API, MCP server and CLI, and what data it collects (including the CLI's install reports and how to turn them off). Both are also in Brazilian Portuguese at the same URLs with `?lang=pt-BR`: [Termos de Uso](https://app.agnthub.sh/terms-of-service?lang=pt-BR), [Política de Privacidade](https://app.agnthub.sh/privacy?lang=pt-BR).