Collapsing list, count, detail and test into one `read_connections` tool keeps the context cost down without losing capability — and the read/write split means you can reason about what an agent is able to change by looking at which half it has been given. The live event stream on the HTTP transport is the other half of the story: a long replication job reports progress rather than going quiet until it finishes.
A management interface for a running CData Sync instance. It exposes the product's REST API as a compact set of tools so an assistant can create connections, run replication jobs and read the execution history.
- Twenty consolidated tools, paired as read and write per entity type, each taking an `action` parameter — so `read_connections` covers listing, counting, detail and connection testing in one tool
- Connections: `read_connections`, `write_connections`, plus `get_connection_tables` and `get_table_columns` for schema discovery
- Jobs: `read_jobs` for status, history and logs, `write_jobs`, `execute_job` to run one immediately, `cancel_job` to stop one, and `execute_query` for custom SQL
- Tasks and transformations: `read_tasks`, `write_tasks`, `read_transformations`, `write_transformations`
- Operations: `read_users`, `write_users`, `read_requests`, `write_requests`, `read_history`, `read_certificates`, `write_certificates`, `configure_sync_server`
- Over HTTP, a Server-Sent Events stream carries live notifications for tool execution, job progress, configuration changes and errors
Node.js 18+ and a running CData Sync instance to point at. Clone, install and build; published as cdata-sync-mcp-server, version 1.0.0. CDATA_BASE_URL and CDATA_AUTH_TOKEN are the connection settings, with CDATA_WORKSPACE optional to scope operations to one workspace. MCP_TRANSPORT_MODE takes stdio, http, or both at once. Docker, Docker Compose, Kubernetes and systemd deployment paths are all documented.
