Sessions & Activity

Sessions are connected MCP clients. Activity is the log of every tool call those sessions have made. Together they answer "who has access, and what have they done?"

The Sessions and Activity views are part of the Agents workspace in the macOS app. The same data is available to AI clients via the mindwiki_list_agent_sessions and mindwiki_get_agent_activity MCP tools.

Sessions

A session is created the first time an AI client authenticates against your vault and persists across that client's MCP requests until you revoke access.

What a session shows

Each session card displays:

  • Client name — what the AI client identifies as (e.g. Claude Desktop, ChatGPT MCP, a custom client name)
  • Protocol version — the MCP protocol version the client speaks
  • Auth method — OAuth or legacy bearer key
  • Status — connected / inactive
  • Started at — when the session was first established
  • Last activity at — most recent tool call
  • Activity count — how many tool calls this session has made
  • Files touched — count of distinct vault files the session has read or written

Click any session for a full activity log scoped to that session.

Connecting and disconnecting

Sessions are created automatically when an AI client makes its first authenticated MCP request. You don't manually create them.

To revoke a session, revoke its underlying credential:

Once the credential is gone, the session immediately stops authenticating. Existing activity records are preserved for audit purposes.

Activity

Activity is the chronological log of every MCP tool call. Whether the call came from Claude reading a page, ChatGPT searching your vault, or a custom script capturing a thought — it's recorded.

What an activity entry shows

Each entry includes:

  • Tool name — which MCP tool was called (e.g. mindwiki_search, mindwiki_write_page)
  • Status — ok or error
  • Duration — milliseconds the call took
  • Arguments — the parameters the AI client passed in
  • Result summary — a short description of what came back
  • Files read — list of vault paths the call accessed (read tools)
  • Files written — list of vault paths the call modified (write tools)
  • Timestamp — when the call happened
  • Session — which AI client made the call

Activity for write tools (mindwiki_write_page, mindwiki_capture, mindwiki_update_index, mindwiki_create_proposal) is especially useful — those are the calls that change content, and you should be able to retrace any AI-driven edit.

How long activity is kept

Activity is retained as long as the underlying database row is — there's no aggressive purge. The macOS app shows the most recent ~50–200 events depending on the view, with deeper history available via the API.

Querying activity from your AI

Any connected AI client can call mindwiki_get_agent_activity to see the recent log. This is useful for prompts like:

Show me everything you've written to my MindWiki vault this week and summarize.

The AI sees the same activity log you see in the macOS view.

REST access

bash
curl -H "Authorization: Bearer $MINDWIKI_API_KEY" \
  "https://api.mindwiki.io/vault/agents/summary"

The summary endpoint returns recent activity, sessions, pending proposals, and vault health in one call. Requires read scope. See API Access.

What this is good for

  • Audit. Know exactly what every AI client has done. No black-box behavior.
  • Debugging. When an AI's response seems off, the activity log shows the exact tool calls and arguments.
  • Permission tightening. If you see a session calling tools you didn't expect, revoke it and re-mint a key with narrower scopes.
  • Self-coaching. Looking at how you actually use AI against your vault often surfaces patterns — the queries you re-ask, the pages you re-read — that suggest where to add structure.

Where to go next