🗺️ Hermes Atlas MCP Server

MCP server for the Hermes Atlas ecosystem directory — gives AI agents instant access to 169+ quality-filtered tools, skills, plugins, and integrations for Hermes Agent.

Why?

Hermes Agent has a massive and growing ecosystem. This MCP server turns that ecosystem into instant expandability — agents can discover, compare, and recommend tools without leaving their conversation.

Quick Start

Add to your MCP client config:

{
  "mcpServers": {
    "hermes-atlas": {
      "command": "npx",
      "args": ["hermes-atlas-mcp"]
    }
  }
}

Or with Docker/stdio:

{
  "mcpServers": {
    "hermes-atlas": {
      "command": "node",
      "args": ["/path/to/hermes-atlas-mcp/dist/index.js"]
    }
  }
}

Tools

Tool	Description	Example
search_repos	Full-text search across 169 repos	search_repos("memory persistence")
list_categories	Browse 12 ecosystem categories	list_categories()
get_repo	Detailed repo info + AI summary	get_repo("NousResearch/hermes-agent")
recommend	Match tools to your use case	recommend("I need to deploy on K8s")
get_featured	Trending/rising repos this week	get_featured()
get_lists	Curated lists overview	get_lists()
get_list	Specific curated list with per-repo descriptions	get_list("best-memory-providers")
ecosystem_stats	Aggregate stats, category breakdown, latest version	ecosystem_stats()
ask_atlas	RAG over research knowledge base (requires embeddings)	ask_atlas("How do skills work?")

Optional: Local Embeddings

The ask_atlas tool provides RAG-powered answers grounded in 27 research files (6,500+ chunks) covering Hermes Agent installation, architecture, skills system, deployment, and best practices.

Install the embeddings (~70MB) separately:

npx hermes-atlas-mcp install-embeddings
# or equivalently:
npx hermes-atlas-install

The server auto-detects the embeddings at startup and adds the ask_atlas tool when available. Without embeddings, all other tools work perfectly using the summaries index.

How It Works

┌────────────────────────────────────────────┐
│           hermes-atlas-mcp                │
│                                            │
│  ┌──────────┐  ┌──────────────┐           │
│  │ repos    │  │ summaries    │           │
│  │ (169)    │  │ (AI-generated│  ← bundled │
│  └────┬─────┘  │  per-repo)  │  or fetched│
│       │        └──────────────┘           │
│  ┌────┴──────────────────────┐             │
│  │ lists, featured, stats   │ ← cached    │
│  └───────────────────────────┘   (4hr TTL) │
│                                            │
│  ┌───────────────────────────┐  optional   │
│  │ chunks.json (70MB)        │ ← install   │
│  │ RAG knowledge base        │   separately│
│  └───────────────────────────┘             │
└────────────────────────────────────────────┘
          │
    stdio (MCP)

• Zero-config: Works immediately with no API keys needed
• Offline-capable: Bundled data works without network; fresh data fetched in background
• Light: Core data is ~300KB; embeddings are opt-in at 70MB
• Fast: Full-text search and recommendations complete in <50ms

Data Sources

All data sourced from ksimback/hermes-ecosystem — a community-curated directory security-reviewed before inclusion.

File	Size	Content
repos.json	60KB	169 repos — owner, name, description, stars, category, official flag
summaries.json	189KB	AI-generated summaries + highlights per repo
lists.json	2KB	6 curated lists (best memory, top skills, deployment, etc.)
list-summaries.json	26KB	Per-repo descriptions within each curated list
featured.json	241B	Currently featured/trending repos
latest-release.json	374B	Latest Hermes Agent version
chunks.json	70MB	6,554 research chunks with pre-computed embeddings (optional)

Development

git clone https://github.com/your-user/hermes-atlas-mcp.git
cd hermes-atlas-mcp
npm install
npm run build

# Test interactively
echo '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"test","version":"0.1"}}}' | npm start

# Watch mode
npm run dev

License

MIT. Data sourced from hermes-ecosystem (MIT/CC BY 4.0).