Handover from Claude to LibreChat
A job started in Claude rarely has to finish there. A Zambo HANDOVER moves the job itself to LibreChat as one checkable record: the goal, the finished steps with a verifiable receipt for each, the decisions, and the exact next step. The receiver in LibreChat starts from the record, not from a summary of it. HANDOVER is Zambo's flagship product.
Before you start: Zambo on the Claude side
In Claude on a Pro or Max plan, open Customize, then Connectors, select Add custom connector, name it Zambo, paste https://zambo.dev/mcp as the remote MCP server URL, and select Connect. There is no key and no local install: Zambo is a hosted remote server.
Full setup: Add Zambo to Claude. The free tier is 20 free calls per day. Work run through Zambo in Claude saves a verifiable receipt per call as it runs, and those receipts are what the handover carries.
Step 1: finish the Claude side and create the handover
- Do the work in Claude through Zambo tools, so each finished step has its receipt saved at the moment it ran.
- At the end of the Claude session, give one wrap up instruction: hand this job over with the goal, what is done, what is next, every decision, and the receipts attached. Zambo assembles the handover as a single record and returns a link in the pattern https://zambo.dev/handover/ followed by the handover ID. Send that one link to whoever continues the job in LibreChat. Nothing else needs to be copied across: no recap, no pasted chat history.
- Open the returned handover link yourself before sending it on. Check the goal reads the way the job was commissioned, the next step is exact, and any step nobody verified is named as not checked.
Step 2: connect Zambo on the LibreChat side
Add a zambo entry under mcpServers in librechat.yaml with type set to streamable-http and url set to https://zambo.dev/mcp, then restart LibreChat: file based changes only initialize at startup. Servers added in the MCP Settings panel take effect without a restart. In a chat, select zambo in the MCP Servers section of the composer's + palette so it shows as a chip, and use a tool compatible model.
Full setup: Add Zambo to LibreChat. New to Zambo on either side? The bridge page walks the whole install: Add Zambo to your AI.
Step 3: open the handover in LibreChat and continue
- Open the handover link in your browser first and read the record: the goal, the finished steps with their receipts, the decisions, and the exact next step.
- In LibreChat, select zambo in the MCP Servers section of the composer's + palette so it shows as a chip in this chat, and confirm a tool compatible model is selected. An entry with chatMenu set to false is kept out of regular chat selection by design.
- Paste the handover link into the chat and say: open this handover, check the receipts that carry weight, record accept or flag on the handover, then continue from the recorded next step.
What travels in the handover
- The goal. The job in the words it was commissioned in, so LibreChat continues the same job and not a retelling of it.
- The state. Each finished step, with the verifiable receipt saved when that step ran in Claude: the tool that ran, the recorded inputs, and the saved result.
- The receipts. One public receipt link per finished step, checkable by anyone with no account. Receipts from a real Zambo session, the shape that travels:
- Receipt d81aeb41: capability_search: the session finds the right tool first
- Receipt 26b353a9: live_price: the BTC price, with CoinGecko named as its source
- Receipt 1f56fe3e: journal_log: the finding written into the job record
- The decisions. Every decision made along the way, with its reason, so the receiver does not quietly re-argue what is already settled.
- The next step. Stated exactly, before the transfer, so momentum survives the switch from Claude to LibreChat.
- What was not checked. Sources or steps nobody verified are named as not checked, not smoothed over.
Verify note
Before accepting, open one or two receipt links from inside the handover and paste each receipt ID into zambo.dev/verify. A match confirms the saved result was not changed since it was saved. That is what a receipt proves: the record is intact, not that the work was right. Judging the work is the receiver's job, now done against the record instead of a recap. Then record accept, or flag what is wrong, on the handover itself. The transfer is not complete until that decision exists.
Why not just paste a summary into LibreChat?
A summary is narration: it can skip the step that failed and drop the decision that constrained everything after it, and none of that is detectable later. The handover record fixes each step under its receipt at the time it ran in Claude, so LibreChat can recompute any step instead of believing it. The general pattern is on the flagship page: AI agent handover.
More handover recipes
- Handover from LibreChat to Claude
- Handover from Claude to ChatGPT
- Handover from Claude to Goose
- Handover from Open WebUI to LibreChat
- Handover from Claude to dots (ChatGPT)
- Handover from Claude to Claude Duet
- Handover from Claude to Meta Muse
- Handover from Claude to Open WebUI
- Handover from ChatGPT to LibreChat
- Handover from Claude Duet to LibreChat
- Handover from dots (ChatGPT) to LibreChat
- Handover from Meta Muse to LibreChat
- Handover from Goose to LibreChat
- All handover recipes
A handover is only as strong as the receipt the receiving side can check. The AER-1 badge kit carries that same verifiable-receipt proof outward, as an embeddable badge for the pages and profiles where the work is shown.