MCP project memory
Give every MCP client the same project memory.
Palimpse is a standard MCP server that carries the state of your project: decisions with their why, open tasks, versioned documents and session history. Connect it once, and Claude, ChatGPT, Cursor or any compatible client picks the project up where it stands.
Every session starts blank. Memory servers don’t fix that.
MCP gave agents hands: they can read your files, query your services, act on your tools. Context is still the missing piece. Each new conversation opens with no idea of what the project is, what was already tried, or what happened yesterday.
Memory servers solve part of it. They store facts and preferences an agent can recall later. But a fact store is not project state. It does not tell the agent which decision superseded which, why an approach was ruled out, or which open task blocks the next step.
A project context should include the decisions and their rationale, the work still open, the documents that carry durable knowledge, and the history of how the project changed over time.
AI memory helps an agent remember facts. Project continuity helps a project preserve its state, history and rationale so that someone or something else can pick it back up later.
Connect it like any MCP server.
Palimpse speaks standard MCP over OAuth. No config files, no tokens to paste: authorization happens in the browser.
Create a workspace
Sign up and create a project for the work you want to keep continuous. Solo is free.
Paste the URL
In your MCP client, add Palimpse as a remote server using the /api/mcp URL from your workspace settings. claude.ai, Claude Code, ChatGPT and Cursor all accept it in their connector or MCP settings.
Authorize in the browser
A browser window opens, you approve access, and the client is connected. Each teammate connects with their own account, so every session is attributed to the human behind it.
Briefed at the start. Logged at the end.
When a session starts, the agent calls catch_up and gets the current state of the project. Open tasks come first, then what changed since its last visit, then the durable facts the project has learned.
Open tasks
Move full-text search to websearch_to_tsquery: agents write boolean queries that silently return zero results. Reconcile the positioning doc with the new home.
Since your last visit
Two sessions. The home was rebuilt around project continuity, then updated when the proposal doc moved to four families.
Latest decision
The simulated chat demo was ruled out: the conversation is not the product. What survives the conversation is.
Project memory
UI copy is English, direct voice. The design stays monochrome.
During the session, the agent works with normal tools: it reads, writes and searches documents, records decisions with the alternatives that were rejected (a new decision can supersede an earlier one), and opens or closes tasks that stay visible until they are actually done.
When the session ends, log_session writes the record: a summary, the decisions made, the documents touched, the tasks opened or closed. Every document version keeps its provenance: who wrote it, from which tool, in which session.
Decisions preserve the why. Open work preserves the present. Documents preserve the knowledge. Sessions preserve the movement.
The same project, from any client.
Palimpse is not tied to one assistant. Connect it from the tools you already use, and they all read and extend the same history.
MCP is the channel, not the point. What travels through it is continuity. Project continuity is the ability to resume a project with its current state, history and rationale intact, regardless of which human, AI, tool or session picks it up next.
Connect your first project in minutes.
Paste the /api/mcp URL, authorize in the browser, and every session starts briefed. Free for solo. Team is free during V1.