Agents
MCP
Streamable HTTP at /mcp. Compact payloads, since_minutes, and a playbook in server instructions.
With the API running, agents talk Streamable HTTP at http://127.0.0.1:4318/mcp.
The server is agent-first: compact payloads, since_minutes, and a playbook in server instructions. Tools still call the same execute services as HTTP. Do not attach a raw SQLite MCP.
Cursor
.cursor/mcp.json:
{
"mcpServers": {
"localtrace": {
"url": "http://127.0.0.1:4318/mcp"
}
}
}Claude Code
.mcp.json in the project you are debugging:
{
"mcpServers": {
"localtrace": {
"type": "http",
"url": "http://127.0.0.1:4318/mcp"
}
}
}Inspector
npx @modelcontextprotocol/inspector http://127.0.0.1:4318/mcpTools
| Tool | What the agent gets |
|---|---|
list_facets | Valid trace filter values |
list_traces | Recent traces (paginated). Prefer since_minutes |
get_trace | Overview (default) or detail=full |
get_span | One span with attributes |
get_trace_sql | DB queries in a trace, sorted by duration |
get_trace_spans | Typed payloads: sql / redis / mongo / prisma / http / express / s3 / openrouter / trpc / error |
search_spans | Cross-trace search (name/attributes, type, service) |
get_trace_logs | Logs correlated to a trace_id |
list_log_facets | Services and severity buckets for log filters |
list_logs | Recent logs (attributes omitted unless raw) |
list_services | Services that have ingested traces |
list_metric_facets | Metric names and services |
query_metrics | Recent metric points |
Prompts: investigate_trace, debug_errors, find_slow.
See the investigation playbook for how agents should call these tools.