MCP server
Connect an AI client to a workspace through the Model Context Protocol, with the person's own permissions.
POST https://<workspace>.zmle.com/api/mcp
Endpoint
POST https://<workspace>.zmle.com/api/mcp: Streamable HTTP, JSON-RPC 2.0, protocol versions 2025-11-25, 2025-06-18, 2025-03-26. The server is stateless and off until the workspace switches it on.
OAuth 2.1 with PKCE and dynamic client registration, discovered from /.well-known/oauth-protected-resource. Each connection acts as one person, with that person's permissions and branches. Scope mcp:read reads; mcp:write may only create drafts.
Read tools
sales_summarybusiness_metriclist_recordsreceivablespayablesprofit_and_losscash_positiontax_summarystock_statusemployee_summarypayroll_summaryteam_todaytill_sessionspurchases_summarycustomer_summarystays_summaryrentals_summaryforecastsearch_recordsget_recordDraft tools (need the write scope and the workspace's permission)
create_draft_quotationcreate_draft_invoicecreate_taskLimits: 60 calls per connection per minute, 3000 per workspace per hour, 30 drafts per hour, 64 KB per request.
Connect a client
- An administrator switches AI connections on in Admin → Integrations & API → AI connections (MCP)
- In your AI client, add a custom MCP connector with the address https://<workspace>.zmle.com/api/mcp
- The client opens this workspace's sign-in: sign in and allow the connection
- Clients without OAuth can use a personal token created on the same page, sent as a Bearer header
End a connection
On the AI connections page, End connection stops a client at once. Switching AI connections off refuses every call immediately.
Next: API keys and scopes
What a key may read and write, and the limits it carries from the person who created it.