Every tool calls the same service code as the REST API with the same request context, so the MCP server can never do anything the API cannot. Tenant isolation, role-based field masking and stage rules apply unchanged.
Tools
Tools are task-shaped: reads bundle what an assistant needs to answer in one call; writes are explicit and narrow. Each tool requires the scope listed; a connection without it still sees the tool, and calling it returns a plain error explaining which scope to approve. Tools that send something (submit_deal, request_documents, send_merchant_portal) say so in their descriptions, so assistants confirm with you first.
Read tools
Write tools
The server’s
instructions describe the object model (businesses → people → deals → submissions → offers → advances) and flag which tools have side effects, so assistants orient without trial calls.
Example prompts
Once connected, prompts like these work as-is:- “Show me deals stuck in Missing docs for more than 5 days and draft follow-ups for each.”
- “Which lenders should I send deal ACME Plumbing to, and why?”
- “Send the merchant portal link to the owner of ACME Plumbing.”
- “Someone is calling from +1 212 555 0123. Who is it and where is their deal?”
- “Summarise the bank statements on the Rivera deal: revenue trend, NSFs, open positions.”
- “Which advances are renewal-ready this week? List the merchant, lender and paid-in percentage.”
- “Log an offer from Northwind on deal 3c9d… : $70,000 at 1.32 over 9 months, daily payments.”
- “What is in the pipeline right now, by stage?”
Connect
Connect from Claude
Add the server as a custom connector.
Connect from ChatGPT
Add it through developer mode connectors.