Keep project decisions between Claude Code sessions

What you will do

Choose a real decision and the module it affects, ask Claude Code to save the reason and its limits as project knowledge, then look it up in a new conversation. Saving is deliberate; Lumvise does not automatically preserve every conversation.

Before you connect

• Install Lumvise from the current macOS release. The published installers are for Apple Silicon Macs running macOS 14 or later. Community is the headless edition; Full also includes the desktop workspace and one-click agent integrations.
• Install and configure Claude Code for the project you want to work on.
• Import and index that project in Lumvise before asking the client to retrieve its knowledge.

Connect Claude Code

In Lumvise Full, open Settings → Agent integrations, choose Claude Code, and select Install. The current website setup flow describes this as a one-click, user-scoped local MCP integration.

Community users and manual configurations can follow the current MCP setup instructions. Configure the Lumvise executable and an absolute project path as documented for the selected client. Client configuration formats differ, so preserve the command and arguments when adapting the example.

Both editions provide project knowledge through the local MCP server. In the current website setup implementation, the one-click MCP server has no fixed project; tool requests identify the project root. For manual setup, check that the configured project path is the same project you imported and indexed.

Save one decision beside its source

First ask Claude Code to find the real module or function that owns the behavior. Review its explanation and confirm the decision with your team before saving it. Then ask it to create a decision record attached to that source and read the saved record back.

Illustrative prompt, not a recorded product result: “Find the module that owns [the behavior we discussed]. Summarize the decision we confirmed, including why we chose it and its limits. Save that decision beside the module, then read the saved record back with its source reference.”

The public Lumvise documentation describes creating typed knowledge against a semantic element returned by the tools; the client should use the real returned element identifier rather than guessing one from a path. The actual saved content and source reference depend on the project and decision you choose.

Check it in a later session

Start a new Claude Code conversation for the same imported project and ask for the decision by its subject or source. Confirm that the answer includes the saved reason, limits, and source it belongs to. If you restarted the runtime, reconnect the MCP client before checking again.

Illustrative retrieval prompt: “Find the saved explanation for [the module or function]. Identify the source it belongs to and summarize the decision, its reason, and its limits before suggesting a change.”

These examples have not been run end to end against a fresh Lumvise and Claude Code session for this guide. Treat the prompts as a starting point and verify the returned saved record and source reference in your own project.

If Claude Code cannot find the knowledge

• Check that the right project is imported and indexed in Lumvise.
• Confirm the MCP client is connected to the intended Lumvise installation.
• For manual setup, compare the configured absolute project path with the imported project.
• Check that the Knowledge tools are available to the connected client.
• Use the actual semantic element returned by search when saving a source-linked record.

Know what stays local

Lumvise’s MCP server runs locally and stores project knowledge with the Lumvise installation. Claude Code and its configured model may still process the prompts and responses you send through that client. Local MCP transport does not mean all AI processing stays on your device. Review the current Lumvise privacy notice, your client’s data settings, and the public setup documentation.