# TRACE MCP server

> Provides a standardized audit trail for AI-assisted research workflows by recording tool calls, decisions, and actor attribution.

- Repository: https://github.com/Thru-Echoes/TRACE
- Page: https://hostedmcp.org/servers/thru-echoes/trace
- Transport: —
- Stars: 26
- License: Apache-2.0

## Environment variables

| Name | Required | Secret | Description |
|---|---|---|---|
| `TRACE_FLEET_ROOTS` | no | no |  |
| `TRACE_PROJECT` | no | no |  |

## Tools (17)

- `trace_start_session`
- `trace_end_session`
- `trace_log_tool_call`
- `trace_log_annotation`
- `trace_log_contribution`
- `trace_log_state_change`
- `trace_propose_decision`
- `trace_resolve_decision`
- `trace_get_session`
- `trace_get_events`
- `trace_get_decisions`
- `trace_get_decision_chain`
- `trace_search`
- `trace_export`
- `trace_list_sessions`
- `trace_project_summary`
- `trace_health_check`

## FAQ

### What does TRACE record?

It records tool calls, decisions, annotations, contributions, and actor attribution, including who proposed what and who accepted or revised it.

### How does TRACE differ from existing AI observability stacks?

Unlike tools that capture call-level traces, TRACE captures decision-level provenance, such as who proposed a step and what alternatives were rejected.

### Where is the session data stored?

Data is stored in self-contained, human-readable, and git-diffable JSON files located in ~/.trace/sessions/.

### Is TRACE available on PyPI?

No, the TRACE server is not on PyPI; users should install it from a repository clone or via a pinned VCS source.
