What it does
Data Prism sits between LLM agents or MCP clients and internal APIs that hold customer data. It pseudonymises personal data per privacy scope, refuses data that has not been classified, and gives each entity one consistent synthetic identity across sources within a scope. When source systems disagree about an entity, the output reports the disagreement instead of hiding it. Two MCP tools ship today: get_entity_context and compare_entity_sources.
It suits Java/Spring platform and backend teams that place LLM agents or MCP clients in front of internal APIs holding personal data. Pseudonyms differ between investigations, so records do not correlate across cases by accident. An optional hash-chained audit trail, enabled with dataprism.audit.sink: hash-chained, is available from version 0.3.0. The standalone server is the primary deployment option, and a Spring Boot starter is available for embedded use.
The standalone server runs with an OAuth2 resource server, parallel mTLS connectors and an embedded Hazelcast identity cache. A Docker Compose quickstart starts the standalone server, a synthetic fixture API and a local HTTPS JWT issuer so a get_entity_context call can be tested end to end. Connecting your own agent client or API requires configuration in YAML and access to the relevant systems.
Limitations and requirements
Pseudonymised data is still personal data under GDPR Art. 4(5). Sending its output to a third-party model still requires a lawful basis, a DPIA and, where the provider is outside the EU, a transfer mechanism. The read budget is shared only across members that have joined one cluster. The audit verifier cannot detect truncation of a writer's most recent records or deletion of a whole process boot's records, and the trail does not resist anyone who already has write access to the file. Only two of four planned MCP tools are built. The re-identification operator surface and the Elasticsearch connector and its search tools are not built.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 17 Nov 2026 · its page on the source
Data Prism
Fail-closed privacy layer that pseudonymises enterprise API data for LLM agents and MCP clients.
v0.8.0As shown on registry.modelcontextprotocol.io when we last read it