MCP Monitoring & Operations

This skill provides instructions for identifying, monitoring, and debugging MCP servers in the AtlasTrinity system.

1. Enumerating Active MCP Servers

To list all currently active and connected MCP servers, DO NOT search for manual command line tools. Instead, use the dedicated health check script:

bash

python scripts/check_mcp_health.py

This script will output:

•A list of all configured MCP servers.
•Their connection status (ONLINE/OFFLINE).
•Latency (ping).
•A summary of overall system health.

Example Output

code

  ✓ react-devtools         T3     ONLINE        1       168ms
  ✓ googlemaps             T3     ONLINE        1       120ms
  ...
  ✓ Summary: 19 online, 0 degraded, 0 offline (of 19 total)
  ✓ Health: 100%

2. Inspecting MCP Configuration

To see the raw configuration of MCP servers (what command launches them, their environment variables, etc.):

bash

cat config/mcp_servers.json.template
# OR (if deployed)
cat ~/.config/atlastrinity/mcp/config.json

3. Debugging Connection Issues

If a server is reported OFFLINE in step 1:

•Check the logs for that specific server in ~/.config/atlastrinity/logs/ (e.g., googlemaps_server.log if available, or brain.log).
•Verify the server's binary or script path exists.
•Check scripts/check_mcp_health.py --verbose (if supported) or just run the health check again to see if it was transient.

4. Verification

When verifying steps related to MCP server logic:

•Always run python scripts/check_mcp_health.py to confirm the server is actually reachable.
•Do NOT assume a server is running just because the process exists; the connection must be established.