Atropos Health announced on June 11 that its Evidence Agent is now available as a Model Context Protocol (MCP) server on the Databricks Marketplace, giving AI agents built on Databricks direct access to Alexandria, a library of over 33 million precision evidence-based findings (pEBFs) derived from real-world patient data. The launch expands a partnership between the two companies first announced in June 2025.

Atropos Health will showcase the integration at the Databricks Data + AI Summit in San Francisco, June 15-18.

How It Works

The Evidence Agent MCP surfaces Alexandria’s clinical evidence as a native capability inside Databricks AI Playground and customer-built agents. Clinicians, researchers, and life sciences teams can query the MCP to get evidence-backed answers drawn from real-world patient outcomes without leaving their Databricks environment, according to Atropos Health’s announcement.

The agent uses multiple LLMs orchestrated together, with each result evaluated for relevance and evidence quality. According to Atropos Health’s product page, the system achieves an 88% “Answered with Evidence” success rate when Alexandria’s proprietary content is combined with published literature, and every response includes a quality assessment badge and rationale for clinician trust.

Part of a Broader Databricks Healthcare Push

Atropos is one of several healthcare MCP providers launching on the Databricks Marketplace simultaneously. Databricks’ own blog post announced MCP servers from Climb (covering PubMed, ClinicalTrials.gov, OpenFDA, and biomedical ontologies), Kythera Labs (medical semantics), and Redox (clinical data pipeline integrations) alongside the Atropos listing.

All MCP servers are governed under Databricks’ Unity Catalog, which means access controls, audit trails, and data governance apply uniformly to the clinical evidence flowing through agents. For regulated industries where data provenance and auditability are non-negotiable, this centralized governance layer is the difference between a proof of concept and a production deployment.

The Agent Evidence Problem

Healthcare is a sector where AI hallucination carries direct patient safety risk. The standard approach of letting a general-purpose LLM generate clinical recommendations from its training data creates liability and accuracy problems that health systems are unwilling to accept at scale.

Atropos Health’s approach inverts that model. Instead of relying on what an LLM “knows,” the Evidence Agent retrieves specific findings from a curated evidence library and grounds its responses in documented real-world outcomes. Alexandria’s 33 million pEBFs are generated through Atropos’s GENEVA OS platform and undergo a multi-layered evidence review process, according to the company’s published methodology.

The MCP standard itself matters here. By packaging evidence retrieval as an MCP server, Atropos makes the capability portable across any MCP-compatible agent framework, not just Databricks. The Databricks Marketplace listing is the distribution channel, but the underlying protocol means the same evidence source could plug into agents running on other platforms that support MCP.

Atropos Health was named to Fierce Healthcare’s “Fierce 15” list for 2026 and has an existing collaboration with Microsoft Dragon Copilot at Stanford Medicine, where its Evidence Agent delivers AI-powered evidence within physician workflows.