MCP

Zebflow exposes a full MCP (Model Context Protocol) server per project. AI agents can manage pipelines, templates, docs, and data through MCP tools.

Endpoint

POST /api/projects/{owner}/{project}/mcp

Requires Authorization: Bearer {token} and Accept: application/json, text/event-stream.

curl -H "Cookie: zebflow_session=superadmin" \
  http://localhost:10610/api/projects/superadmin/default/mcp/session

Returns the session token for MCP calls.

Orientation — start_here, version, help, help_search
Pipelines — pipeline_register, pipeline_activate, pipeline_list, pipeline_execute, etc.
Templates — template_write, template_edit, template_list, template_get, etc.
Docs — docs_project_read, docs_project_write, docs_agent_read, docs_agent_write
Data — connection_list, connection_describe, credential_list
Resources — move_resource, install_ui_components, list_ui_catalog
Git — git_command