MCP & Agent tools

Architecture pattern. To maximize reasoning and agentic interoperability, K Pro integrates with customer tooling through MCP configurations rather than bespoke per-vendor connectors. Customer systems (e.g. ELN, LIMS, data lake, BI) are connected to K Pro as MCP servers, either Owkin-provided wrappers for common tools or the customer's own servers. Authentication and per-tool authorization flow through OIDC/SAML or service accounts/API keys (especially for bulk data querying where it may be more efficient).

Data stays in customer systems; K Pro queries in place. New integrations are MCP-server additions.

Representative integrations and complementary functionality:

• Benchling, Dotmatics (ELN/LIMS). Pull assay and entity data on demand into a reasoning session; push predictions and annotated records back through the same MCP surface.
• Snowflake, S3, Databricks, GCS. Read-only querying with customer credentials; no duplication into K Pro. The customer may run data processing components to augment their data with Owkin and industry standard models, or render it more compatible with agentic reasoning.
• KNIME, Pipeline Pilot. K Pro callable from REST or Python nodes as a pipeline step.
• Spotfire, Tableau. K Pro writes structured outputs back to the customer warehouse; existing dashboards refresh against those tables.