No approval needed
Always availableHow your AI finds out what it may do, and what to tell you when it may do nothing.
describe_connection
Assist With Moving keeps the durable move record. An AI you choose can reason in its own environment, use bounded tools, and save useful work back to the move. Signing in proves who you are; a grant you approve decides what it may do.
Human part
Connect an AI you choose
Create a private move.
Leave a durable Queue handoff.
Sign in from a compatible AI, then approve a grant.
Review saved work inside the move.
Sign-in ties calls to your account, the grant decides what may be done, and revoking it refuses the very next call.
Create a private move and leave a Queue handoff first. Then paste the endpoint below into an AI that speaks remote Streamable HTTP MCP with compatible OAuth, sign in, and approve a grant for the moves and operations you actually want covered.
https://movingmanifest.com/mcp
Open https://movingmanifest.com/ai and help me use Assist With Moving. If you speak remote Streamable HTTP MCP with compatible OAuth, connect to https://movingmanifest.com/mcp. Signing in is not permission — I then approve a grant that decides what you may do. Call describe_connection if no tools appear, get_move_brief first, search before creating duplicates, and save finished work with save_complete_result.
Signing in proves who you are. A separate approval decides what your AI may do — which moves, which data, which operations. Revoking it stops the very next call.
When you approve evidence reading, your AI can open the private photos for that work, create or correct inventory, keep source notes, and mark what still needs review.
Your AI reads a compact move brief, searches before creating duplicates, and adds or corrects the move, place, and inventory context you approved.
One call saves the decisions, estimates, readable plan result, and honest source checks together — and can close the Queue handoff in the same approval.
Your AI is shown only the tools your grant covers. Everything else is not merely refused — it is not listed at all. One tool, describe_connection, is always there so your AI can tell you what it is missing instead of failing silently.
How your AI finds out what it may do, and what to tell you when it may do nothing.
describe_connection
Route, dates, rooms and places, belongings and boxes, decisions, estimates, saved results, source checks, and Queue summaries — for the moves you select.
get_move_brief, search_move_records, get_move_records
Returned through Assist With Moving as protected content, never as a storage link. Reading context does not include this.
get_evidence_media
Move context, inventory, decisions, estimates, plan sections, source checks, and one complete reviewed result. Replay-safe, so a retry corrects instead of duplicating.
save_move_context, save_inventory, save_planning_record, save_complete_result
See what is waiting, take it, ask you the smallest question, hand it back, or mark it Done with its result attached.
list_queue_work, claim_queue_work, release_queue_work, ask_queue_question, complete_queue_work
Reversible archive and restore. You can put any of it back, and nothing is ever permanently deleted.
archive_move_records
We would rather tell you what is unfinished than let a page imply a connection that nobody has completed end to end yet.
The endpoint, the five approvals, the grant boundary, the Queue workflow, and immediate revocation are built. No AI product has yet completed a full connect, approve, read, save, revoke, and reconnect run against it, and the provider-side setup for verified client identity is not finished. Any AI that speaks remote Streamable HTTP MCP with compatible OAuth can attempt it; that is a requirement, not a promise about a particular product.
Settings → AI connections is where you approve a grant, choose all moves or only some, read the consent summary, see each connection's client, last use, and activity, and revoke it. Until you approve one, a connected AI sees only describe_connection, which tells it to ask you.
The separate API-key catalog is available today for headless or non-OAuth tools. Create a key only for an AI you trust, then revoke it when it is no longer needed.
No connection required. Open the Queue item, copy its bounded brief into any AI, and paste the result back into the same item.
The others exist, and they are not equivalent. Their tool catalogs are different from each other, so a tool working in one place says nothing about another.
Sign-in plus the grant you approve. The tool list on this page, and nothing outside it.
Keeps connections that were made through the older gateway working. A different and larger catalog, governed by that older gateway rather than by grants. Not the same door and not the same tools.
Takes a scoped mmk_ key and never offers sign-in. Its own separate catalog for headless or non-OAuth tools.
Runs on your own machine with the same kind of scoped key, serving the API-key catalog rather than the canonical one.
Create your private move and keep only the context that is useful now.
Leave a Queue handoff in your own words when work should survive the chat.
Paste the endpoint into an AI that speaks remote MCP with OAuth, and sign in.
Approve a grant: choose the moves and the operations it covers.
Ask it to save finished work, then review the result inside the move.
Open the Queue item in your move and copy its bounded brief into any AI, however you normally work. Paste the result back into the same Queue item and mark it Done. Nothing about this path needs MCP, OAuth, or a key, and the handoff still keeps its scope, instructions, evidence references, and result.
A client can hold on to an old tool list. If a tool that used to work starts failing, or a new one never appears, disconnect the connection and connect again. Your saved move work is untouched, and your grant stays as you approved it.