MCP Guide

MCP Guide: session_info Tools

This skill uses two MCP tools from the session_info server to access today's Cowork sessions.

Tool 1: mcp__session_info__list_sessions

Purpose: Returns a list of all Claude Cowork sessions.

Call pattern:

mcp__session_info__list_sessions()

No required parameters. The tool returns session metadata including session IDs, timestamps, and titles/descriptions where available.

Date filtering: The tool may return sessions across multiple days. Filter the results by comparing the session timestamp against today's date (from currentDate in your context: 2026-06-05 format). Keep only sessions where the date portion matches today.

What the response includes (typical fields):

• session_id or id — use this to call read_transcript
• created_at or timestamp — use for date filtering and ordering
• title or description — may be available, may be null
• message_count — helpful to know session size before reading

If the call fails: Tell the user "I couldn't connect to the session index right now — try again in a moment" and stop.

Tool 2: mcp__session_info__read_transcript

Purpose: Returns the full message history for a given session.

Call pattern:

mcp__session_info__read_transcript(session_id: "<id>")

Pass the session_id (or id) from the list_sessions response.

Response format: Typically returns an array of messages with role (user/assistant) and content fields.

Reading strategy for large transcripts:

• If a session has many messages, focus on: the first user message (what they asked for), the last few assistant messages (what was delivered), and any messages containing words like "here's", "I've created", "saved to", "next steps", "TODO", "follow up"
• You don't need to read every line — extract the signal, skip the noise
• For very long transcripts (50+ messages), skim by reading every 3rd-5th exchange

Ordering: Process sessions newest-first so the report opens with the most recent work.

Error Handling

Example Flow

1. list_sessions() → returns 4 sessions
2. Filter to today → 3 sessions match
3. Order by timestamp descending → [session_C, session_B, session_A]
4. read_transcript(session_C) → most recent session
5. read_transcript(session_B) → second session  
6. read_transcript(session_A) → earliest session
7. Extract + synthesize → write report

Notes

• The session_info MCP is read-only — this skill never modifies or deletes anything
• Session data is private to the user's Cowork workspace
• If list_sessions returns sessions without timestamps, try to infer order from session IDs or content clues

Asana Tools

This skill uses two Asana MCP tools to pull today's task activity.

Tool 3: mcp__555be8f8-37df-4a20-8299-f9d35d9f3e69__asana_search_tasks

Purpose: Find tasks modified or created today across the user's Asana workspace.

Getting the workspace GID first: Call mcp__555be8f8-37df-4a20-8299-f9d35d9f3e69__asana_list_workspaces with no arguments to get the workspace GID. Use the first workspace returned (Patrick has one workspace: Enterprise Platform).

Call pattern for today's modified tasks:

asana_search_tasks(
  workspace: "<workspace_gid>",
  modified_since: "<today's date in ISO format, e.g. 2026-06-06T00:00:00Z>"
)

Also useful: To find tasks completed today, add completed: true and completed_since = today's date.

What the response includes:

• gid — task ID, use this for get_stories_for_task
• name — task title
• completed / completed_at — whether it was finished
• created_at — whether it's a new task
• modified_at — last modified time
• assignee — who owns it
• projects — which project(s) it belongs to

Limit results: If many tasks are returned, focus on tasks where modified_at is today (not just the search boundary).

Tool 4: mcp__555be8f8-37df-4a20-8299-f9d35d9f3e69__asana_get_stories_for_task

Purpose: Get comments and activity log for a specific task.

Call pattern:

asana_get_stories_for_task(task_gid: "<task_gid>")

What the response includes:

• type — "comment" (user-written) or "system" (auto-generated activity like status changes)
• text — content of the comment or activity description
• created_at — when it was added
• created_by — who added it

Filter to today: Only surface stories where created_at is today's date.

Don't over-fetch: Only call get_stories for tasks that appear interesting (completed today, have recent comments, or were flagged as blockers). Skip routine system activity like "moved to section".

Updated Error Handling (Asana)

Updated Example Flow

1. list_sessions() → 3 sessions today
2. read_transcript(session_C, B, A) → extract Claude work
3. list_workspaces() → Enterprise Platform workspace GID
4. search_tasks(workspace, modified_since=today) → 8 tasks
5. get_stories_for_task(task_1) → 2 comments today
6. get_stories_for_task(task_4) → 1 comment today
7. [skip tasks with no comments/stories]
8. Synthesize everything → write unified report