Gemini CLI

Connect Crow to Gemini CLI, Google's AI assistant for the terminal.

Prerequisites

• Node.js 18 or later
• Crow cloned and set up locally (for stdio) or a deployed gateway (for remote)
• Gemini CLI installed (npm install -g @google/gemini-cli)

Option A: Local (stdio)

Best for development — runs Crow servers directly on your machine. No gateway or network required.

Setup Steps

1. Clone and set up Crow:

   git clone https://github.com/kh0pper/crow.git
   cd crow
   npm run setup

2. Edit ~/.gemini/settings.json and add Crow servers under mcpServers:

   {
     "mcpServers": {
       "crow-memory": {
         "command": "node",
         "args": ["/path/to/crow/servers/memory/index.js"],
         "env": {
           "CROW_DB_PATH": "/path/to/crow/data/crow.db"
         }
       },
       "crow-projects": {
         "command": "node",
         "args": ["/path/to/crow/servers/research/index.js"],
         "env": {
           "CROW_DB_PATH": "/path/to/crow/data/crow.db"
         }
       },
       "crow-sharing": {
         "command": "node",
         "args": ["/path/to/crow/servers/sharing/index.js"],
         "env": {
           "CROW_DB_PATH": "/path/to/crow/data/crow.db"
         }
       },
       "crow-blog": {
         "command": "node",
         "args": ["/path/to/crow/servers/blog/index.js"],
         "env": {
           "CROW_DB_PATH": "/path/to/crow/data/crow.db"
         }
       },
       "crow-storage": {
         "command": "node",
         "args": ["/path/to/crow/servers/storage/index.js"],
         "env": {
           "CROW_DB_PATH": "/path/to/crow/data/crow.db",
           "MINIO_ENDPOINT": "localhost",
           "MINIO_PORT": "9000",
           "MINIO_ACCESS_KEY": "your-access-key",
           "MINIO_SECRET_KEY": "your-secret-key"
         }
       }
     }
   }

   Replace /path/to/crow with the absolute path where you cloned Crow. Omit crow-storage if you are not running MinIO.

3. Restart Gemini CLI — it will detect the MCP servers automatically.

Transport

• Type: stdio
• Auth: None (local process)

Combined server (lighter footprint)

If you prefer a single entry point rather than five separate servers, use the crow-core combined server. It starts with memory tools active and loads other servers on demand:

{
  "mcpServers": {
    "crow-core": {
      "command": "node",
      "args": ["/path/to/crow/servers/core/index.js"],
      "env": {
        "CROW_DB_PATH": "/path/to/crow/data/crow.db"
      }
    }
  }
}

Or generate the config automatically:

cd /path/to/crow
npm run mcp-config -- --combined

Then copy the crow-core entry from the generated .mcp.json into ~/.gemini/settings.json.

Option B: Gateway (HTTP)

Connect to a deployed Crow gateway for remote access — useful for Tailscale setups or cloud deployments.

Prerequisites

• Crow gateway deployed and reachable (Cloud Deploy Guide or Tailscale Setup)

Setup Steps

1. Edit ~/.gemini/settings.json:

   {
     "mcpServers": {
       "crow-memory": {
         "url": "https://your-crow-server/memory/mcp"
       },
       "crow-projects": {
         "url": "https://your-crow-server/projects/mcp"
       },
       "crow-tools": {
         "url": "https://your-crow-server/tools/mcp"
       }
     }
   }

   For a Tailscale-accessible gateway, use the Tailscale address instead:

   {
     "mcpServers": {
       "crow": {
         "url": "http://100.x.x.x:3001/router/mcp"
       }
     }
   }

2. Start Gemini CLI — it will auto-discover OAuth metadata and prompt for authorization on first use.

Transport

• Type: Streamable HTTP
• Protocol: 2025-03-26
• Auth: OAuth 2.1 (automatic discovery)

Verification

Start Gemini CLI and ask:

Store a memory that Gemini CLI is connected to Crow.

Then verify it was saved:

Search my memories for "Gemini".

Cross-Platform Context

Crow automatically delivers behavioral context when Gemini CLI connects via MCP — memory protocols, session management, and transparency rules are active from the first message.

For more detailed guidance, ask Gemini to use MCP prompts: session-start, crow-guide (accepts platform: "gemini" argument), project-guide, blog-guide, or sharing-guide.

Memories and projects stored via Gemini CLI are immediately available on all other connected platforms. See the Cross-Platform Guide.

Tips

• Gemini CLI looks for ~/.gemini/settings.json globally — there is no per-project config file like Claude Code's .mcp.json
• Run npm run mcp-config in the Crow directory to generate a full config, then copy the relevant entries into settings.json
• The crow-storage server requires MinIO; omit it if you are not using file storage