# Tasktrail

> Tasktrail is a private work log for software engineers. AI coding agents (Claude Code, Codex, and any Model Context Protocol client) write to it automatically as you work, recording progress, decisions, blockers, debugging findings, and next steps, organized by project and task. It produces deterministic daily summaries and a web dashboard.

## What it is

- A remote MCP server with its own OAuth, plus a web dashboard.
- Traces are written by your AI agents automatically over MCP (no prompting, no copy-paste), each tagged with its source (e.g. claude-code, codex).
- Every entry is timestamped and typed: progress, decision, blocker, debugging, next_step, general, or summary.
- `summarize_day` produces a deterministic summary grouped by project and task, built from the real log rather than generated by a model.
- Private per Google account; OAuth access tokens are stored hashed at rest and can be revoked at any time.
- Free to start.

## MCP tools

list_workspaces, list_projects, find_tasks, list_tasks, get_task, list_traces, search_traces, get_daily_context, summarize_day, set_current_workspace, set_current_project, create_task, update_task, merge_tasks, create_trace, update_trace, delete_trace. Read tools require the work:read scope; write tools require work:write.

summarize_day is read-only. To persist an end-of-session summary, call summarize_day, then save the returned text with create_trace using type summary.

Every create_trace call must include task_id, including summary traces. Tasks are long-lived and span days or weeks: before creating one, call find_tasks with a few keywords and reuse a matching task's id; only call create_task when nothing matches, and pass the returned task.id. Pass task_id to list_traces to read one task's complete history across all dates. Task titles should be clean, user-friendly outcome phrases. Tasks can include effort: quick, easy, moderate, hard, or major. Missing effort defaults to moderate.

## Connect

- Claude desktop or web app (custom connector): Settings → Connectors → Add custom connector → paste `https://tasktrail.ai/mcp`, then sign in with Google.
- Claude Code: `claude mcp add --transport http tasktrail https://tasktrail.ai/mcp`
- Codex: `codex mcp add tasktrail --url https://tasktrail.ai/mcp` (or add `[mcp_servers.tasktrail]` with `url = "https://tasktrail.ai/mcp"` to `~/.codex/config.toml` by hand)
- Codex plugin: from the repo root, `codex plugin marketplace add .`, then `codex plugin add tasktrail@tasktrail`.

## Links

- Home: https://tasktrail.ai
- Docs: https://tasktrail.ai/docs
- Privacy: https://tasktrail.ai/privacy
- MCP endpoint: https://tasktrail.ai/mcp