How MCP works

How your AI reads company context and follows its evidence through Pensieve.

MCP connects the AI you already use to Pensieve's company context. To install it, go to Setup & clients.

AI clients
Claude
Cursor
ChatGPT
Claude Code
Codex
Windsurf

MCP server

Your data
Pensieve
One context layer, available to the AI clients your team uses.

How your agent uses it

  1. Get oriented. Read the context's root overview to understand the company and where subjects belong.
  2. Read the relevant knowledge. Use maintained Pages for an explanation; search for a specific question, then open the relevant Page or source passage. The tree outline is useful for understanding the map's organisation, rather than as a route to every fact.
  3. Follow the evidence. Open cited sources for exact details. Use the agent's other tools for live records or actions the context layer points to.

The context-layer primer explains Pages, sources and citations. Alongside your tools explains how this works with personal memory and your other integrations.

What your AI can do

Your agent can search and read company knowledge, inspect processing status and history, save new evidence, and edit Pages with your account's permissions. The MCP reference lists the tools and explains how custom clients choose a context.

How writes stay safe

Page writes check the revision they are changing and record the result in history. Client confirmation depends on the client. See Saving data or editing pages for choosing the right destination, attribution, locks and write safeguards.

Connect your client

Install Pensieve for your AI client, sign in and start a new conversation. Ask it to brief you on your company, then follow a citation to check the evidence behind an answer.

Up next

Save and edit