What can AI do with your work history through MCP?
Resume coding, find a product you liked, prepare a client call, or organize a day trip. Try four distinct ways to use ContextBerg records with your usual AI.
Read in Japanese →CONTEXTBERG × YOUR AI
Your day, back in context.
Interactive UI illustration · sample data · no live AI connectionMCP retrieval details
get_activity({"date":"2026-11-11"})
get_agent_history({"date":"2026-11-11","project":"settings-demo","max_sessions":5,"max_turns":5})Illustrative calls · sample dataCursor aligned the save and load paths. The work report records three passing tests, but the restart check in the signed app is still open.
Next check
Open the current diff, change a setting, quit the app completely, and relaunch it. Confirm the value survives before marking the bug fixed.
Connect ChatGPT
Enable the relevant capture and integrations, and keep ContextBerg running. Mac examples require Node.js.
In a desktop environment with MCP settings, add a STDIO server: command npx; args -y @contextberg/mcp-server. ChatGPT web needs a separate connection architecture.
Restart and check /mcp. Retrieved records are passed to your AI client.
Recover the context behind work and everyday plans
Which checks remained after yesterday’s code change? Why did you like that desk? What did you promise to bring to the client call? Useful context is spread across browsing, app activity, and conversations with AI.
MCP lets an AI client use external tools and information. ContextBerg’s local MCP server exposes available work reports and supported agent conversations. The interactive examples above let you inspect how those records could support a response.
Four situations, four different kinds of help
| Task | Records to consult | Useful result |
|---|---|---|
| Resume coding | Editor activity, agent conversations, and a work report | The last change and an outstanding verification step |
| Find a product you liked | Product-page history and an AI comparison conversation | The original page, your reasons, and an unresolved constraint |
| Prepare a client call | Reference documents, planning conversations, and an active deck | What you promised to bring and questions for the agenda |
| Organize a weekend trip | Route-page history and a planning conversation | An unfinished preparation checklist |
Start with the fragment you remember: the desk with cable storage, tomorrow’s Northstar call, or the trip you planned over lunch. A useful response may be a verification step, a recovered preference, a call agenda, or a packing checklist.
Inspect the records behind the answer
A browser visit locates a product page; a saved conversation can explain why you liked it. Opening a proposal does not establish that it was sent, and discussing a travel checklist does not make the errands complete. Inspect the source records behind the answer.
The demo uses sample data to explain the workflow. In the product, available context depends on enabled capture, supported integrations, and the requested period.
Connect your usual AI client
Choose a client in the demo and open its setup instructions for a Mac example. Claude Code and Codex can run the local MCP server. The ChatGPT example targets desktop environments with MCP settings; ChatGPT web requires a separate connection architecture.
Retrieved records are passed to the AI client and its configured model. Check capture scope and model destination before using work information.