Model Context Protocol

Use the Hermes corpus inside Claude, ChatGPT, Grok, DeepSeek or the Hermes Agent.

We expose the entire podcast-transcript search as an MCP server. Plug it into your own AI assistant and it can search every episode, ask the transcripts questions, and pull KPIs — right inside your chat, grounded only in what was actually said.

One URL, any client

It's a hosted, public, read-only streamable-HTTP MCP server. Paste the URL below into any MCP-capable assistant — no Python, no keys, no config file.

Works with

Claude · ChatGPT · Grok · DeepSeek · Hermes Agent · Cursor — and anything else that speaks MCP over HTTP.

01The one thing you need — the server URL

The server is hosted. Whichever assistant you use, this is the remote MCP endpoint you point it at. Works on macOS, Windows, web & mobile; nothing to install.

https://mcp.howtousehermes.com/mcp

Public, read-only corpus — no login or API key required. Below is exactly where that URL goes in each assistant.

02Connect your assistant

Same URL everywhere. Pick your client.

🟣 Claude Desktop / Web

Settings → Connectors → Add custom connector. Paste the URL, click Add. The tools appear under the 🔌 menu in any chat. (Custom connectors need a paid Claude plan.)

🟢 ChatGPT Plus / Pro / Business

Settings → Connectors → enable Developer mode, then Create / Add a connector and paste the URL as a remote MCP server. Also usable from Codex and deep research. (Custom MCP connectors require a paid plan; availability is rolling out by region.)

⚫ Grok xAI

In Grok's Settings → Connectors / Integrations, add a custom MCP server with the URL. If your Grok build doesn't show a connector box yet, point any MCP-capable client (below) or the xAI API's MCP tool at the same URL.

🔵 DeepSeek

The DeepSeek chat app has no built-in connector box today, so connect through any MCP client — Cursor, Cline, Continue, or the DeepSeek API's tool-use — pointed at this same URL. (When DeepSeek ships a native connector, the URL is all you'll need.)

⚡ Hermes Agent

Add it under mcp_servers in ~/.hermes/config.yaml (see step 04). Tools auto-register on startup as mcp_podcast_search_*.

🧩 Cursor & other MCP clients

Add it as a remote / streamable-HTTP server with the same URL. No command, no args — just the URL.

Consumer-app connector menus for Grok and DeepSeek are still rolling out and their exact wording may differ from your build — the endpoint itself is a standard streamable-HTTP MCP server, so the universal fallback (any MCP client / the vendor's API MCP tool) always works.

03What you get

Seven tools, all returning clean JSON. Episode results include a clickable URL.

search_transcripts — full-text search across every indexed episode (sort, min-relevance, limit).
ask_episodes — a natural-language question answered ONLY from the transcripts, with cited sources.
get_stats — corpus size & freshness (episodes, strong matches, last update).
recent_episodes — the most recently added episodes, with any KPIs they report.
list_themes — trending themes/keywords that recur alongside Hermes.
get_episode — full metadata (and optionally the transcript) for one episode.
daily_digest — short per-day bullet summaries of each day's new episodes (≤10 bullets/day).

04Add to Claude Desktop (manual / local fallback)

Only needed if you prefer running the server locally instead of the one-click URL above. Config lives at ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows). You need Python with the mcp SDK: pip install "mcp>=1.0".

{
  "mcpServers": {
    "hermes-podcast-search": {
      "command": "python",
      "args": ["/path/to/mcp_server.py"],
      "env": {
        "PODCAST_API_BASE": "https://howtousehermes.com"
      }
    }
  }
}

The server is a thin, read-only HTTP client over this site's public API — it needs no database and no secrets, so it runs anywhere. Grab mcp_server.py from the project (or ask the site owner). Fully restart Claude Desktop after editing the config.

05Add to the Hermes Agent

Drop this under mcp_servers in ~/.hermes/config.yaml and restart the agent. Tools register as mcp_podcast_search_* and are available in every conversation.

mcp_servers:
  podcast_search:
    command: python3
    args:
      - /path/to/mcp_server.py
    env:
      PODCAST_API_BASE: "https://howtousehermes.com"
    timeout: 90
    connect_timeout: 30

Prefer the hosted server? Instead of the local script above, point Hermes at the remote URL: url: https://mcp.howtousehermes.com/mcp (with timeout/connect_timeout) — no local Python needed.

06Try it

Once connected, just ask your assistant things like:

“Search the Hermes podcasts for subagents and summarize what people say.”
“Using the podcast transcripts, how are people using Hermes for SEO, and what results did they get?”
“What are the newest episodes, and what KPIs do they mention?”