Setup manual · agents over MCP
Wire the journal into your agents.
Tasktrail is one OAuth-protected remote MCP endpoint. Point a compatible agent at it, approve access with Google once, and your work starts keeping its own trace.
https://tasktrail.ai/mcpChoose your instrument
Every client speaks the same protocol; only the wiring differs. Flip to yours.
Claude Desktop + Web
No terminal required — add Tasktrail as a custom connector and sign in once.
- 01
Open Settings → Connectors in Claude Desktop or claude.ai
- 02
Choose Add custom connector
- 03
Name it Tasktrail and paste the endpoint
- 04
Connect with Google and approve the work log scopes
https://tasktrail.ai/mcpClaude opens the Tasktrail OAuth flow. After approving, call get_daily_context to confirm the connector is active.
Custom connectors require a paid Claude plan. On the free plan, use instead — it works on any plan.
Teach it what to keep
Connecting grants the tools; one standing instruction sets the editorial line. Add this note to your project instructions (CLAUDE.md, AGENTS.md, or your client's custom instructions) so agents log what matters and skip the noise.
Use Tasktrail automatically throughout the session. At session start, call get_daily_context. When needed, call list_workspaces and set_current_workspace. When the project is clear, call set_current_project. Use create_task/update_task for real units of work. Before create_task, call find_tasks and reuse a matching task's id; reopen done tasks instead of duplicating them. Task titles should be clean, user-friendly outcome phrases. Every create_trace call must include task_id, including summary traces; when needed, call create_task first and pass the returned task.id. Use task effort when useful: quick, easy, moderate, hard, or major. Use create_trace for meaningful progress, decisions, blockers, debugging findings, and next steps. Keep trace content to one or two sentences; put deeper detail in technical_details. When a trace corresponds to a commit, PR, or issue, include it in links. Do not log trivial file reads, routine commands, or minor observations. Before re-deriving a past solution or decision, call search_traces first. At session end, call summarize_day. If a persistent summary is useful, save it with create_trace using type summary.
The tool kit
Seventeen tools, two pouches. Reads answer questions; writes land only in your private workspace.
get_daily_contextlist_workspaceslist_projectslist_tasksfind_tasksget_tasklist_tracessearch_tracessummarize_day
set_current_workspaceset_current_projectcreate_taskupdate_taskmerge_taskscreate_traceupdate_tracedelete_trace
When something jams
Four fixes cover nearly every stuck connection.
01Tools do not appear in the client
Make sure the OAuth flow finished: in Claude Code, run claude mcp login tasktrail; on claude.ai, check that Settings → Connectors shows Tasktrail as connected (custom connectors require a paid Claude plan). Then start a new conversation so the tool list reloads, and call get_daily_context to test.
02401 or “unauthorized” after a while
Access tokens last one hour and clients refresh them automatically with a 30-day refresh token. If the refresh token has expired or been revoked, re-authenticate: run claude mcp login tasktrail in Claude Code, or disconnect and reconnect the Tasktrail connector on claude.ai.
03Revoking or reconnecting
Remove the Tasktrail connector under Settings → Connectors on claude.ai, or remove the server from your client's config; revoked tokens are rejected with a 401 on the next call. To reconnect, add the connector again and approve access.
04Not sure traces are landing?
Sign in at Tasktrail— new traces appear in the dashboard's day feed as they are logged, tagged with their source (e.g. claude-code).