Connect Claude to your workspace
Pick where you'll use LeadMove, Claude, ChatGPT, Claude Code, Cursor or another tool, and choose what it may do.
Settings → Developers → Connect an assistant, then pick where you'll use it: Claude, ChatGPT, Claude Code, Cursor or another tool. Each path shows only what it needs.
You sign in with your normal LeadMove account, tick the permissions you want, and the assistant can then answer questions about your pipelines, buyers and leads from your own data. Nothing is granted until you approve it on the LeadMove screen, and you can revoke it in one click afterwards.
This page covers Claude and Claude Code. ChatGPT and Cursor are in other MCP clients; a script, or a tool that only takes a header, uses a key.
Claude.ai and Claude Desktop
Pick Claude in Connect an assistant. The dialog opens Claude's connector settings and has the URL ready to copy. No key is involved.
- In Claude, open Settings → Connectors → Add custom connector.
- Paste
https://app.leadmove.io/mcp. - Sign in. The LeadMove screen shows which application is asking, which account it will act as, and one checkbox per permission. Claude does not ask for specific permissions, so only Read starts ticked: tick everything you want it to do now, because permissions can be removed from a connection later but never added. Name the connection after where it runs — "Claude on my laptop" — because that name is what you will read later in Settings and in the activity log.
- Connect.
Leave the LeadMove dialog open while you do this: it waits for the connection and says Connected with its name as soon as you approve.
On a Team or Enterprise plan, an Owner adds the connector once under Organization settings → Connectors; members then connect individually with their own account.
Claude Code
Pick Claude Code in Connect an assistant for this command, ready to copy:
Run it in your terminal, then type /mcp, pick leadmove and sign in. A browser opens on the same permission screen, and the dialog in LeadMove turns to Connected.
On a machine with no browser, a server or a CI job, choose No browser on that machine? Use a key instead under the command: connect with a key.
The permission prompts
Claude asks you before it uses a LeadMove tool, and its prompt names the tool: Pipelines, Preview: pause a buyer, Apply a previewed change. A tool titled Preview: only shows you what would change, a tool named after what it shows only reads, and Apply a previewed change is the one that makes the change you just read. Approve every LeadMove tool once or always, except Apply a previewed change, which you approve each time. Nothing changes until that one runs.
What you see in Claude
In Claude.ai and Claude Desktop, Claude shows LeadMove's preview card: what would change, who is affected, whether it can be undone. You decide on the card itself: Apply makes the change, Discard throws the preview away, and once applied, Undo puts it back for 7 days. Telling Claude yes works just as well.
Claude may ask you to allow the action before it runs; that box is Claude's own. The card then says what happened (applied, undone, discarded, expired, or replaced by a newer preview when something changed in the meantime) and links to the change in your activity log. It keeps LeadMove's colors in Claude's light or dark theme and counts down its last two minutes.
Claude doesn't see clicks on the card. To carry on from there, press Tell Claude: it puts a line such as "I applied it from the card: Pause BrightHome Energy." in your message box, and nothing is sent until you send it.
A preview that only simulates routing has no Apply on the card: say yes to Claude, which reads you the result. Claude Code and other clients show the same preview as text.
A whole setup (a pipeline, its buyers, their delivery methods and rules, proposed together) is one card: one line per item with its state, the first four shown and Show more for the rest. Create all creates everything, paused or switched off; afterwards Undo all takes all of it back for 7 days.
Previews kept for later have a card too. Ask Claude what is waiting and each kept proposal shows with the date it is kept until, an amber line if your data changed since it was drafted, and Review in LeadMove, which opens it in Settings → Developers to check against today's data and apply. You can also ask Claude for the same change again.
Come back to a conversation days later and its cards show what happened since: applied, undone, discarded or expired, whoever did it, from the card, from Claude or from the app. A card says Checking… for a moment while it asks. A preview still inside its ten minutes keeps its buttons.
Slash commands with the LeadMove plugin
The plugin is the same connection with the six workflows attached as slash commands, so you type /leadmove:triage instead of picking it out of /mcp.
Then /mcp and sign in, as above. The permission screen is the same one; installing a plugin grants nothing by itself.
| Command | What it does |
|---|---|
/leadmove:triage | What needs a human right now, grouped by cause |
/leadmove:weekly-report | The week against the week before. Pass client for the shareable version |
/leadmove:audit | What is wired wrong or about to break, worst first |
/leadmove:fix-failed | Groups yesterday's failures by cause and proposes one fix per group |
/leadmove:onboard-buyer | An insertion order becomes a buyer, its endpoint and its rules, in one preview |
/leadmove:setup-pipeline | A description becomes a pipeline, its sources and its rules, in one preview |
Each takes optional arguments: /leadmove:triage "since Friday" solar-ca.
The command files are generated from the server's own workflow definitions, so a workflow reads the same whether you invoke it as a slash command or from /mcp. The repository is public: github.com/leadmove/claude-plugin.
/plugin uninstall leadmove@leadmove removes the commands. It does not revoke the connection: do that in Settings.
Choosing permissions
| Permission | What it allows | Role needed |
|---|---|---|
| Read your workspace | See pipelines, buyers, rules, leads, delivery traces and reports | Member |
| Change configuration | Create and change buyers, endpoints, rules and pipelines, after you approve each change | Admin |
| Act on leads | Approve, reject, send and retry leads, after you approve | Admin |
| Record buyer payments | Record a buyer's payment on their prepaid balance, after you approve | Owner |
A connection can never do what the person behind it could not do by hand. Permissions above your role are greyed out, and if your role changes later the connection narrows with it, on its next request.
Read is enough for almost everything people ask first: "why wasn't this lead delivered", "which buyers are near their cap", "what changed on this pipeline last week".
Permissions can be removed from a connection at any time, never added. To widen one, connect again — that way every grant is a dated, named event rather than a silent upgrade to a credential that is already out there.
Disconnecting
Settings → Developers → Connected assistants, open the connection, then:
- Suspend — a pause. Nothing works until you resume it. Use this while you investigate something.
- Revoke — permanent. The next request fails immediately, including one already in flight. There is no un-revoke; reconnect instead.
Above the list, four switches turn a whole permission off for every connection at once. That is the fastest stop: you do not have to know which connection to revoke. See what a connected assistant can and cannot do.
Questions
Do I need a plan upgrade? No. Connecting an assistant is available on every plan, and connections don't count against your lead quota.
Can a member connect? Yes, with read access only, and their connection always sees contact details masked.
Why does Claude ask me to sign in again? Access tokens are short-lived and refresh silently; a fresh sign-in means the connection was revoked, suspended, or left untouched long enough for its refresh token to lapse. Check the row in Settings before reconnecting — if it says Suspended, resuming it is the fix.
What does it see? Your organization's data, and nothing from anyone else's. See data sent to a connected assistant.