MCP Access: Connect Your Companion to Other AI Tools

What You'll Need

• 🔑 A Generated TokenAn MCP token from your companion's Integrations page on Eidolon (except ChatGPT — see below).
• 🤖 A Supported AI ToolClaude Desktop, Claude Code, ChatGPT, Hermes, or Codex.
• 🌐 The Server URLhttps://mcp.geteidolon.app

Step 1: Generate a Token

You'll need one token per AI tool you want to connect. Each token authenticates to a single companion.

1. Open your companion's profile page on Eidolon.
2. Scroll to the Integrations section.
3. Click Generate Token, enter a label (e.g. "Claude Desktop"), and click Generate.
4. Copy the token immediately — it's shown only once. It looks like: eid_a1b2c3d4e5f6a1b2c3d4e5f6a1b2c3d4e5f6a1b2

Step 2: Configure Your AI Tool

Claude Desktop

Edit your Claude Desktop configuration file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "eidolon": {
      "url": "https://mcp.geteidolon.app",
      "type": "http",
      "headers": {
        "X-MCP-Token": "YOUR_TOKEN_HERE"
      }
    }
  }
}

Restart Claude Desktop. MCP tools will appear with the eidolon_ prefix.

Claude Code (CLI)

Add to ~/.claude.json (global) or .mcp.json (per-project):

{
  "mcpServers": {
    "eidolon": {
      "url": "https://mcp.geteidolon.app",
      "type": "http",
      "headers": {
        "X-MCP-Token": "YOUR_TOKEN_HERE"
      }
    }
  }
}

Run claude in your terminal. Tools are available immediately.

ChatGPT (OAuth 2.1)

ChatGPT uses OAuth — no token needed. The setup is fully automatic:

1. In ChatGPT, add a custom MCP server with URL: https://mcp.geteidolon.app
2. ChatGPT discovers our OAuth metadata automatically.
3. You'll be redirected to the Eidolon authorization page to pick your companion and grant read-only access.
4. ChatGPT exchanges an authorization code for a Bearer token — no copying required.

Hermes

Hermes ships in two flavors and each one uses a different config format. Pick the one that matches how you run Hermes:

Hermes CLI — config.yaml

If you run Hermes from the terminal, add an MCP server to your Hermes config.yaml:

mcp_servers:
  eidolon:
    url: https://mcp.geteidolon.app
    headers:
      X-MCP-Token: YOUR_TOKEN_HERE
    connect_timeout: 30

Restart Hermes. Tools are available automatically.

Hermes Desktop App — JSON config

If you run the Hermes desktop app, add the MCP server using a JSON config (not YAML):

{
  "url": "https://mcp.geteidolon.app",
  "connect_timeout": 30,
  "headers": {
    "X-MCP-Token": "YOUR_TOKEN_HERE"
  }
}

Restart the Hermes desktop app. Tools are available automatically.

Codex (CLI)

Add to ~/.codex/mcp.json (global) or .mcp.json (per-project):

{
  "mcpServers": {
    "eidolon": {
      "url": "https://mcp.geteidolon.app",
      "type": "http",
      "headers": {
        "X-MCP-Token": "YOUR_TOKEN_HERE"
      }
    }
  }
}

Codex discovers the tools on next session start.

What Data Is Available

Once connected, your AI tool gets access to 15 read-only tools across these categories:

Step 3: Teach Your Agent How to Use the Tools

After connecting, your AI tool will discover the tools automatically — but it won't know the workflow patterns for using them effectively. Tell your agent to load the skills reference:

Load the Eidolon MCP agent skills as context. Use those tool workflows
when accessing my Eidolon companion's data.

Some tools support loading the skills file directly. Download the Agent Skills Reference and share it with your agent:

Download Agent Skills (Markdown)

The skills doc teaches your agent the recommended workflow patterns — starting with identity confirmation, building context through the journal, and when to use each tool.

Managing Tokens

• List tokens: View all active tokens in the Integrations section
• Revoke: Click the trash icon on any token to immediately disable it
• Last used: See when each token was last accessed
• Token labels: Name each token (e.g. "Claude Desktop", "Work Laptop") to keep track

Security & Privacy

Troubleshooting

FAQ