Empower your healthcare agents with ready-to-use MCP on Databricks Marketplace

Databricks launched ready-to-use Model Context Protocol (MCP) servers on its Marketplace to help healthcare organizations build AI agents with specialized biomedical tools. The servers, developed in partnership with Climb and Atropos Health, provide direct access to curated data sources including PubMed, ClinicalTrials.gov, and real-world evidence libraries. This integration aims to overcome data silos and reduce engineering efforts required for AI adoption in healthcare workflows.

by Yen Low /blog/author/yen-low , Mark Lee /blog/author/mark-lee , Matthew Giglia /blog/author/matthew-giglia , Nicholas Siebenlist /blog/author/nicholas-siebenlist , Jay Bhankharia /blog/author/jay-bhankharia , Paul Ford /blog/author/paul-ford and Itai Weiss /blog/author/itai-weiss While there holds great promise for AI agents to transform the healthcare industry, for agents to be successful, they require a sophisticated integration of curated knowledge, timely data, and specialized tools. This is where the Model Context Protocol MCP https://modelcontextprotocol.io/docs/getting-started/intro as an open standard interface is key to integrating all the above add-ons to supercharge agents with the appropriate capabilities. Building effective AI agents in healthcare presents unique challenges. First, healthcare workflows demand highly specialized and curated knowledge, ranging from genomic data to clinical guidelines, that is difficult to build from scratch. Generic large language models LLM often lack the specialized information needed for drug discovery or clinical decision support. Furthermore, given the higher stakes in healthcare, LLMs need grounded context and direct access to purpose-built tools to provide safe and accurate answers. Traditionally, healthcare data and tools have remained siloed across disparate systems, requiring massive engineering efforts to integrate and creating significant barriers to AI adoption. As such, healthcare professionals continue to labor in paper-heavy and data-intensive workflows, exactly the kind of tasks that agents are designed to simplify and alleviate. Databricks already provides a full-service agent ecosystem https://docs.databricks.com/aws/en/agents/ where it is easy to discover and reuse essential building blocks—including MCP servers, function tools, Genie spaces and Vector Search �—all governed under the robust security of Unity Catalog. We are making it easier than ever to jumpstart development with ready-to-use MCP servers on the Databricks Marketplace https://marketplace.databricks.com/ . With these MCP servers, you can introduce new capabilities like: | | Generate hypotheses from disease-target-drug interactions or look up bioactivity including toxicity: | | | Search for biomedical literature: | Query clinicaltrials.gov to learn about current trials and their status: | | | Examine drug product labels and drug/device safety information: | Query Medicare coverage policies: | | Search ontologies: | | | | | Get answers to clinical questions with proprietary real-world evidence in | | | Design and manage healthcare data integrations using natural language with | We partnered with Climb https://climb.ai/ , a trusted Databricks systems integrator partner, to provide direct access to key biomedical MCP servers, including PubMed, Clinicaltrials.gov and US Census. The diagram above shows how Climb brings public life-sciences data and your own private data together in your Databricks workspace. Canonical public sources are connected through live MCP services, with no copies and ETL to keep in sync. Inside Databricks they meet your private Gold-layer records and your own models, all under the Unity Catalog governance you already run. From surfaces like Genie and Agent Bricks, your teams can query public and private evidence together, powering use cases like responder/non-responder dashboards and adverse-event monitoring, each in a single audited run. Atropos Health https://www.atroposhealth.com/ MCP enables quick access to Alexandria® https://www.atroposhealth.com/alexandria/ , the Atropos Evidence™ Library, a collection of over 33M precision evidence based findings pEPF™ and precision real-world evidence pRWE™ summaries. These observational studies close the evidence gap on real world multi-morbid cohorts that are often underrepresented in clinical trials and other conventional studies. Additionally, it promotes evidence-based medicine and streamlines clinical operations by surfacing evidence relevant to individual patient profiles. "The best partnerships don't just connect two products — they unlock something neither could offer alone. That's what being a Databricks Marketplace partner means to us. By bringing real-world evidence directly into the Data Intelligence Platform, we're giving life sciences and healthcare teams the ability to build AI applications grounded in clinical reality, not just data - Dr. Brigham Hyde, CEO and co-founder of Atropos Health. Kythera Labs https://www.kytheralabs.com/ provides clinical semantic infrastructure for healthcare AI, helping agents translate natural language clinical concepts into the standard medical vocabulary logic required to work with real-world healthcare data. The Kythera Clinical Semantic Translation MCP Server enables users to describe diseases, medications, procedures, laboratory tests, outcomes, and eligibility criteria in plain language and automatically translates them into retrieval-ready code sets across vocabularies such as ICD-10-CM, SNOMED CT, CPT, HCPCS, NDC, RxNorm, and LOINC. By bridging the gap between clinical language and structured healthcare data, Kythera accelerates cohort discovery, patient finding, clinical trial feasibility, commercial analytics, market access research, and real-world evidence generation. Rather than relying on manual code lookups or specialized healthcare data expertise, organizations can use Kythera's MCP server to quickly move from clinical questions to actionable insights, enabling AI agents and analysts to work more effectively across claims, EHR, pharmacy, laboratory, and research datasets. Redox https://redoxengine.com/ is your interoperability partner, powering intelligent healthcare data exchange at scale. Provider, payer, healthtech, EHR, and med tech organizations rely on Redox to connect, prepare, route, and execute complex data workflows across a connected network of more than 12,000 systems. The Redox MCP server https://redoxengine.com/blog/llms-mcps-and-agentic-ai-oh-my/ helps customers manage and monitor real-time healthcare interoperability feeds using natural language. Creating a new integration connection historically required a technical user to navigate multiple screens, or chain together multiple Platform API calls to set up in a repeatable pattern. This blog https://www.databricks.com/blog/months-minutes-building-real-time-clinical-data-pipelines-natural-language shows how nearly any user can use the Redox MCP server https://redoxengine.com/blog/llms-mcps-and-agentic-ai-oh-my/ to set up a new integration, test it, review the output, move it to production, monitor the logs, and review the real-time data flowing into Databricks, all in one session. You can also bring your own MCP servers by adding them to our MCP Catalog https://docs.databricks.com/aws/en/generative-ai/mcp/ . These can be MCP servers externally hosted https://docs.databricks.com/aws/en/generative-ai/mcp/external-mcp-usage or hosted on Databricks Apps https://docs.databricks.com/aws/en/generative-ai/mcp/custom-mcp-usage . Additionally, many Databricks resources like Genie Spaces, AI Search, Unity Catalog functions and SQL Warehouses are also available as Managed MCP servers https://docs.databricks.com/aws/en/generative-ai/mcp/managed-mcp . More crucially, because your data is already on Databricks, you can use it to build bespoke agents to unlock valuable insights from your data assets. All MCP servers, whether sourced from the Databricks Marketplace https://marketplace.databricks.com/ or your own custom builds https://docs.databricks.com/aws/en/generative-ai/mcp/custom-mcp-usage , are centralized in the Databricks MCP Catalog https://docs.databricks.com/aws/en/generative-ai/mcp/ and governed by our Unity AI Gateway https://docs.databricks.com/aws/en/ai-gateway/ to ensure fine-grained, secure access. All MCP servers, whether from an external host or hosted on Databricks, are centralized and governed in the Databricks MCP Catalog for fast discovery and re-usability. Whether you prefer a low-code experience using Agent Bricks https://docs.databricks.com/aws/en/generative-ai/agent-bricks/multi-agent-supervisor or flexible coding options with Databricks Notebooks https://www.databricks.com/product/collaborative-notebooks now with Genie Code https://docs.databricks.com/aws/en/genie-code/ AI assistance , both beginners and power users can easily build and deploy agents to production. Databricks offers full observability of the agent workflows in MLflow traces https://mlflow.org/genai/observability?gad source=1&gad campaignid=23556358454&gbraid=0AAAABCrGXuJ48nsoezgx774UOJ1wBLs1S&gclid=Cj0KCQjwof QBhCgARIsADaMzOdwgXtziCSdUUGUxDxdXEvK42ONvtHbvfScZp zEbfDpTepF4GVM1gaAjaZEALw wcB and facilitates evaluation with built-in AI judges and custom scorers https://docs.databricks.com/aws/en/mlflow3/genai/eval-monitor/concepts/scorers . Additionally, Databricks provides an accompanying Review App https://docs.databricks.com/aws/en/mlflow3/genai/human-feedback/expert-feedback/live-app-testing to collect expert human feedback. Once deployed, the agent is continuously monitored https://mlflow.org/ai-monitoring with AI guardrails https://docs.databricks.com/aws/en/ai-gateway/guardrails to protect against malicious activity. Below we show how you can use these MCP servers, whether you are a beginner, intermediate or advanced Databricks user You can start asking questions like “Give me the molecular properties for orforglipron”. If you have connected the ChEMBL MCP server, it should answer the following: For scenarios involving numerous tools and MCP servers, we suggest building a Supervisor Agent https://docs.databricks.com/aws/en/generative-ai/agent-bricks/multi-agent-supervisor create-a-multi-agent-supervisor-system . This orchestrator efficiently routes queries to various specialized sub-agents, each focused on a particular toolset. You can do this in 5 mins with the no-code Agent Bricks Supervisor Agent https://docs.databricks.com/aws/en/generative-ai/agent-bricks/multi-agent-supervisor create-a-multi-agent-supervisor-system . For deeper insights from MCP results alongside your proprietary information, you can integrate your own data as Genie Spaces or Vector Search https://docs.databricks.com/aws/en/generative-ai/agent-bricks/multi-agent-supervisor supported-subagents-and-tools as additional tools. As an illustration, we connected a DrugBank Genie Space to stand in for a chemical library. This enables us to ask questions like “List the GLP-1 agonists in DrugBank and their ADMET properties”. It should generate the appropriate SQL to get the following answer: Upon deployment, Agent Bricks provides a REST endpoint https://docs.databricks.com/aws/en/generative-ai/agent-bricks/multi-agent-supervisor step-5-query-the-agent-endpoint automatically for easy invocation. For those needing a custom interface, you can develop one within Databricks Apps https://docs.databricks.com/aws/en/dev-tools/databricks-apps/ using a variety of available app templates https://docs.databricks.com/aws/en/generative-ai/agent-framework/chat-app . For full customizability, you can code the supervisor agent using any agent framework e.g. Langgraph, OpenAI Agents in Databricks Notebooks https://www.databricks.com/product/collaborative-notebooks on the cloud or a local IDE https://www.databricks.com/product/data-science/ide-integrations . There are multiple agent and app templates https://github.com/databricks/app-templates/tree/main to choose from so you would hardly need to start from scratch. They even come with Agent Skills https://github.com/databricks/app-templates/blob/main/.claude/AGENTS.md for AI coding assistants like Genie Code https://docs.databricks.com/aws/en/genie-code/ or Claude Code for a faster start. Additionally, these templates work with Supervisor API https://docs.databricks.com/aws/en/generative-ai/agent-framework/supervisor-api-app build-a-custom-agent-using-the-supervisor-api where you can simply define the tools. For example, AiChemy https://www.databricks.com/blog/aichemy-next-generation-agent-mcp-skills-and-custom-data-drug-discovery , our drug research agent, was created using this code repository https://github.com/databricks-industry-solutions/aichemy with configurable MCP servers so you can search external knowledge bases e.g. PubChem along with your own data e.g. ZINC chemical library on AI Search https://www.databricks.com/product/artificial-intelligence/ai-search with ECFP embeddings for structurally similar molecules like this: Whether you are just starting out or are an experienced developer, you can create your own agent using your data and MCP servers in a matter of minutes. With the growing selection of MCP servers on the Databricks Marketplace https://marketplace.databricks.com/ , you can quickly level up your agents. Additionally, to accelerate your data transformation, agent development and MCP server needs, please connect with our trusted Databricks partner, Climb http://climb.ai/ . Climb specializes in operationalizing enterprise AI through production-ready solutions that deliver measurable business outcomes. To learn more details about their newly launched MCP Service for healthcare and life sciences, please read more here http://climb.ai/labs/hls-mcp/ . For partners interested in listing their commercial MCP servers on our growing Databricks Marketplace https://marketplace.databricks.com/ , please review our MCP partner validation portal here https://databrickslabs.github.io/partner-architecture/data-collaboration/mcp-marketplace-validation . Subscribe to our blog and get the latest posts delivered to your inbox.