External integrations
RagBox is an optional service deployed separately. Their inclusion in MCPBox means the application contains the integration code and settings; it does not start those services or install their dependencies.
RagBox: ingestion and retrieval
- Start RagBox and prepare the tenant and credentials for your deployment.
- Open the RagBox integration settings in MCPBox.
- Enable the integration and enter the reachable base URL, tenant ID and API token. Add the admin token if you need administrative collection or vector-connection operations.
- Save, synchronize knowledge bases and connect the required collection to a project.
- Test ingestion and search with a document whose contents you know.
Supported configuration modes:
| Setting | Values |
|---|---|
| Write mode | bleve_only for local indexing; dual_write for local and RagBox ingestion. |
| Search mode | bleve_only, ragbox_only, dual_search (concatenated and sorted results), or hybrid (merged results). |
The default integration mode values are dual_write and hybrid; the integration itself must still be enabled and configured. Collection service settings and bindings also affect which path is used. Check reported errors and results instead of assuming a configured URL means ingestion succeeded.
Local Bleve search does not generate embeddings. External semantic retrieval and vector connections are configured through RagBox. Source files and RagBox data remain owned by their respective service.
Connectivity and diagnostics
Save configuration under System → RagBox, then verify the connection. Catalogs load when needed; an unavailable service shows an error, Retry and technical details in its section. Local Bleve knowledge works without RagBox.
PolicyBox removed
v1.4.0 no longer provides PolicyBox settings, runtime playground or admin policybox commands. Legacy identity-verification flags block affected projects until migration. Service data is retained: PolicyBox migration.