AgentKit Documentation

AgentKit is a curated registry served by a Model Context Protocol (MCP) server: every skill is vendored from a public upstream repository and published with a SHA-256 content hash (verified on every read), its license and its risk flags. Upstream commit pinning is not yet complete; each manifest reports its exact status. See the Security & Trust page for the review model and its limits.

1. Quickstart & Agent Setup #

AgentKit works with any MCP-compatible client including Hermes Agent, Claude Code, and OpenAI Codex.

MCP Server Configuration (Streamable HTTP)
{
  "mcpServers": {
    "agentkit": {
      "url": "https://agentkit.callen.page/mcp"
    }
  }
}

2. Built-in MCP Tools #

search_skills(q, category?, tag?, page?, pageSize?)

Search published skills with filters, pagination, fuzzy matching, and suggestions.

get_skill(name) • batch_get_skills(names) • get_related_skills(name)

Read one skill, fetch up to 10 manifests at once, or find related skills.

list_skill_versions(name) • diff_skill_versions(name, from?, to?)

Inspect published versions and compare version labels for a skill.

get_usage_stats(window?, limit?) • get_service_diagnostics() • get_server_events(limit?)

Usage leaderboards, service health/protocol support, and pollable server events.

get_install_bundle(client) • submit_skill_feedback(kind, skill?, rating?, message)

Exact per-client install bundles and capped, identity-free skill feedback.

Full list: 19 tools, 7 doc resources, 8 prompts — see Websites MCP docs and `tools/list`.

3. REST API Reference #

MethodRouteDescription
POST/mcpMCP server (Streamable HTTP): tools + llms-txt resource. Optional Bearer token for higher caps.
GET/api/skillsPublic skill catalog (same data as search_skills)
GET/api/usage?window=all|day|week&limit=nUsage leaderboard (same data as get_usage_stats)
GET/mcp/documentation-indexSection index of the agent documentation
GET/api/healthService health check

4. Usage & Limits #

Free and open for everyone — no API key needed. Anonymous callers get generous per-IP fair-use limits; signed-in accounts and API keys get higher caps. Limits keep the service fast for all.

LimitAnonymous (no account)Account / API key
Requests200 / min per IP900 / min per IP
Concurrent requests530
Search results20 per search25 per search
PriceFreeFree