In this post, we will cover:
- Why AI agents need direct data access, not just a REST API
- How MCP turns DISGENET into a live catalog of named, auditable tools
- What changes for grounded answers, multi-step reasoning, and audit trails
- Why this matters especially for pharma target identification
- How to connect the DISGENET MCP Connector in 3 steps
For more than a decade, DISGENET has given researchers and pharma teams a way to systematically explore gene-disease associations and scores that can be traced back to their evidence. Today, we’re extending that same trusted data to a new kind of user: the AI agent.
Researchers are increasingly turning to AI agents to help with target identification, variant triage, and literature synthesis. But an agent is only as good as the data grounding it. Without a direct connection to a source like DISGENET, an AI tool falls back on general training data and the open internet, which means surface-level answers, outdated figures, or claims that can’t be traced back to real evidence.
We’ve released the DISGENET MCP Connector, built on the Model Context Protocol (MCP), enabling AI tools to query DISGENET’s disease genomics data directly. Instead of relying on general web knowledge or handwritten API calls, AI agents can autonomously determine which associations, scores, and evidence to retrieve.
Why Not Just Use the REST API?
Our REST API isn’t going anywhere. It remains the right choice for partners with deterministic pipelines, high-throughput bulk data pulls, and performance-critical integrations, and that isn’t changing. To fully take advantage of this tool, developers know exactly which endpoint to call, which parameters to provide, and how to combine multiple requests. An AI agent often doesn’t; it has to figure that out at runtime, and asking it to write raw HTTP requests from documentation alone invites hallucinated endpoints, incorrect or unsupported parameters, invalid identifier formats, and dead ends with no way to self-correct.
MCP closes that gap. Rather than guessing at API calls, an agent queries a live catalog of named tools we’ve built and validated, such as looking up gene-disease associations for a target or pulling variant-level evidence, and our server handles the actual execution.
What MCP Changes
An MCP server is a semantic layer, not a competing API. Instead of an agent guessing at HTTP calls, it queries a live catalog of named tools, things like find_disease_genes, and our server executes the call underneath. That has a few concrete effects:
- Grounded, traceable answers. Every association returned through the connector comes directly from DISGENET’s evidence-backed knowledge base, with links to the underlying supporting sources and associated confidence scores, rather than an unverifiable web summary.
- Multi-step reasoning made simple. An agent can chain steps like “find genes for this disease → check their association scores → cross-reference known variants” in one session, without having to track the state itself.
- Deterministic, auditable execution. The agent never writes a raw request against our data; it selects a named tool, and our own code runs the query. For teams making target prioritization calls with AI assistance, that audit trail matters.
- Built for how teams already work. The connector fits naturally into tools like Claude, enabling rapid prototyping for data scientists exploring disease-gene relationships, and it composes cleanly alongside other MCP-enabled resources —drug and chemical knowledge bases, clinical trial data, structural biology tools as part of a larger agentic workflow.
Why This Matters Especially in Pharma
For teams making target identification or prioritization decisions with AI assistance, the difference between an agent writing a raw API call and an agent selecting a pre-built, audited tool is not a minor implementation detail — it’s a safety and governance question. MCP keeps a human-auditable boundary between “what the model decided” and “what code actually ran against your data.”
Built for the Way Pharma Teams Are Already Working
The MCP connector is designed for:
- AI agents and LLM-based workflows — Claude, GPT-based tools, or custom LLM agents that need to reason autonomously about which DISGENET data to pull
- Rapid prototyping by pharma data scientists working in tools like Claude Desktop or Cursor
- Multi-agent pipelines, where DISGENET sits alongside drug and chemical knowledge bases, clinical, or structural biology tools, all speaking the same protocol
Getting started
The DISGENET MCP Connector is available as a remote MCP server using this URL:
https://mcp.disgenet.com/mcp
To connect:
- Add the DISGENET MCP server to your MCP-compatible AI assistant or application using the server URL above.
- When prompted, sign in to your DISGENET account.
- Once authenticated, you can immediately start querying DISGENET using natural language.
Available for everyone with a DISGENET account