Connecting Clients

Step-by-step instructions for connecting various AI clients to AgentiBridge.

Prerequisites

• AgentiBridge running with SSE transport (AGENTIBRIDGE_TRANSPORT=sse)
• Server accessible at http://HOST:PORT (default: http://localhost:8100)
• API key configured if using authentication (AGENTIBRIDGE_API_KEYS)

Quick check:

curl http://localhost:8100/health
# Should return: {"status": "ok", "service": "agentibridge"}

Claude Code CLI

Add to ~/.mcp.json (or project-level .mcp.json):

{
  "mcpServers": {
    "agentibridge": {
      "url": "http://localhost:8100/sse",
      "headers": {
        "X-API-Key": "your-api-key"
      }
    }
  }
}

Verify:

claude --mcp-debug
# Should show "agentibridge" in the connected servers list

Available Tools in Claude Code

Once connected, Claude Code can use all 16 tools:

• “List my recent sessions” → list_sessions
• “What did I work on yesterday?” → list_sessions with since_hours=24
• “Search for Docker setup sessions” → search_sessions
• “Show me session abc-123” → get_session
• “What tools did I use?” → get_session_actions
• “Find sessions about authentication” → search_semantic
• “Summarize that session” → generate_summary
• “Continue from session abc-123” → restore_session
• “Run this task in the background” → dispatch_task
• “Did that job finish?” → get_dispatch_job

ChatGPT Custom GPT (Actions)

1. Go to ChatGPT → Explore GPTs → Create a GPT
2. In the Configure tab, click Create new action
3. Set:
   • Authentication: API Key
   • API Key: Your AGENTIBRIDGE_API_KEYS value
   • Auth Type: Custom Header
   • Header Name: X-API-Key
4. Import the OpenAPI schema from your server, or manually configure the SSE endpoint URL

Example Action Schema

openapi: 3.0.0
info:
  title: AgentiBridge
  version: 0.2.0
servers:
  - url: http://your-host:8100
paths:
  /health:
    get:
      operationId: healthCheck
      summary: Check service health
      responses:
        '200':
          description: OK

Note: ChatGPT Actions work best with REST endpoints. For full MCP integration, use the SSE endpoint with an MCP-compatible client.

Claude Web (MCP Servers)

Claude.ai supports connecting to remote MCP servers:

1. Go to Settings → MCP Servers → Add Server
2. Enter:
   • Name: AgentiBridge
   • URL: http://your-host:8100/sse
   • Headers: X-API-Key: your-api-key

Grok (xAI)

For Grok or other clients that support MCP:

1. Use the SSE endpoint: http://your-host:8100/sse
2. Set authentication header: X-API-Key: your-api-key

Generic MCP Client

Any MCP client that supports the SSE transport can connect:

from mcp import ClientSession
from mcp.client.sse import sse_client

async with sse_client(
    "http://localhost:8100/sse",
    headers={"X-API-Key": "your-key"},
) as (read, write):
    async with ClientSession(read, write) as session:
        await session.initialize()

        # List available tools
        tools = await session.list_tools()
        for tool in tools.tools:
            print(f"  {tool.name}: {tool.description}")

        # Call a tool
        result = await session.call_tool("list_sessions", {"limit": 5})
        print(result)

Troubleshooting

Connection refused

• Verify the server is running: curl http://HOST:PORT/health
• Check if the port is open: ss -tlnp | grep 8100
• If behind a firewall, ensure port 8100 is allowed

401 Unauthorized

• Check your API key matches one in AGENTIBRIDGE_API_KEYS
• Verify the header name is X-API-Key (case-insensitive)
• Try with ?api_key=your-key as a query parameter

SSE connection drops

• Check reverse proxy settings (buffering must be disabled)
• Increase proxy timeouts
• Verify the server isn’t running out of memory

No sessions found

• Run agentibridge status to check transcript directory
• Trigger collection: call the collect_now tool
• Verify ~/.claude/projects/ contains .jsonl files