Skip to content

AnswerShared project context

One memory for ChatGPT, Claude and Copilot

Central Brain gives supported ChatGPT desktop, Claude and GitHub Copilot clients access to the same project memory through MCP. A decision saved with memory_checkpoint from one connected client can be retrieved with memory_recall from another connected to that project. Your assistants share a local project store; their native account memories and chat histories remain separate.

Reviewed against current product documentation:

A project can move from planning in Claude to implementation in Copilot and review in the ChatGPT desktop app. Repeating the brief at each handoff costs time and makes subtle disagreements harder to spot. Central Brain by NeuroAIgent is built around one inspectable project record that those supported tools can query.

Start with Central Brain pricing and confirm the installer currently available in getting started. Use the client-specific configuration generated by the app. Local client access, supported modes and organization policies vary; an MCP connection working in one desktop client does not establish browser or mobile support.

What is shared between the assistants?

Central Brain keeps decisions, facts, questions, todos, summaries and a project brief in readable local files. Its file search separately indexes prepared project documents. Sharing those records lets a second tool find the same decision and supporting source instead of relying on a summary pasted from another chat.

The shared record is the Central Brain project: save the material you want to reuse, then ask the next assistant to retrieve it explicitly. Your OpenAI, Anthropic and GitHub accounts keep their own preferences and native memory settings. Past conversations enter this durable project record through reviewed checkpoints, rather than a silent whole-history import.

Which connection should each client use?

The official OpenAI MCP documentation describes desktop server settings separately from remote integrations. GitHub documents its own MCP configuration formats. Keep the command and arguments from Central Brain, but use the destination format required by the client instead of copying an entire JSON wrapper blindly.

ClientConnection workflowVerify before handoff
Claude Desktop / Claude CodeUse the Central Brain configuration generated for that client.Ping the project root and inspect the memory tools.
ChatGPT desktop appUse the local MCP setup supported by your app and account. OpenAI documents STDIO servers in desktop settings.Select the intended local connection and confirm a real recall call.
GitHub Copilot in a supported local clientUse its generated MCP entry and client configuration.Confirm local server reachability and agent tool access.

Set up a handoff you can actually verify

Use a harmless test fact first. Verify the tool result in the second assistant: that makes the shared Central Brain record visible and confirms which saved context carried into the new session.

  1. Create one Central Brain project for the engagement. Choose and index only the relevant source folders.
  2. Connect the first client using that project configuration. Call central_brain_ping and inspect the root.
  3. Save a specific test decision, such as an approved document version, with memory_checkpoint. Verify the note contents.
  4. Connect the second client to the same project root. Start a fresh session, ping the root, and ask memory_recall for that decision.
  5. Ask the second client to find a supporting document with semantic_search or search_text and include the returned source reference.
  6. Repeat with the third client. Record the configuration and available tools so the same working setup is easy to repeat.

Keep a clean project record as work changes

Agree on what deserves a checkpoint. A useful decision note says what was agreed, why, and what remains open. Distinguish a user-approved decision from a model suggestion. Include a source path or document version when it helps a later assistant verify the context.

Review the project brief regularly. Central Brain can replace that brief through memory_compact while marking older notes as compacted. Inspect the proposed summary for omitted conditions and obsolete decisions before accepting it. For simultaneous editors or several machines, define coordination and backup explicitly; automatic multi-user conflict resolution is outside this local project workflow.

Understand where retrieved context goes

Source files, prepared output, vectors and memory remain on storage you choose in the default workflow. When a cloud assistant calls a tool, it receives the returned context and processes it under its own provider and workspace settings. Avoid describing that interaction as zero data leaving the computer.

Synced storage, backups, optional cloud connections and device access also affect the boundary. Use a separate project for each client or matter and inspect the root after switching. The security guide explains licensing and connection boundaries. Central Brain makes continuity deliberate: select the project, retrieve the saved record and verify the supporting context.

Make the decision record specific enough to survive a handoff

A reusable checkpoint should say what was agreed, why it was agreed, which source or document version supports it, and what remains unresolved. For example: “Use the approved October brief for the proposal; the delivery date remains open until the client confirms it.” Label a proposed idea as a proposal rather than quietly turning it into an approved fact.

Before changing assistants, inspect the saved note and ask the second client to retrieve it in a fresh session without pasting the answer. Confirm both clients identify the same project root. If the next person uses another computer, agree how approved files are transferred and which version is current; multiple licenses do not themselves establish synchronization or concurrent editing.

  1. Select one project and an owner for its current brief.
  2. Save one reviewed decision and one open question, with the relevant source version.
  3. Open the second supported client and verify the project before recalling either record.
  4. Correct missing or stale notes, then repeat the recall check before proceeding.

Common questions

Which ChatGPT setup uses the local Central Brain connection?

Use the supported desktop MCP setup. A web integration needs its own supported connection architecture; a local executable path alone is not a web endpoint.

How does Copilot use a note saved from Claude?

A supported Copilot client can retrieve that note from the same Central Brain project through MCP. The record stays in Central Brain, separate from native Copilot Memory.

Why did the next assistant miss my decision?

Check that the checkpoint exists, both clients point to the same root, and the new assistant actually called memory_recall. Also check tool permissions and the client connection status.

Next steps

Sources and review notes