Self-Proposed GSoC 2026 Idea: Autonomous Genomic Discovery Agent via MCP

[saurabhhhcodes]

🏗 cBioPortal: The Autonomous Genomic Agent

Architecture Diagram

graph LR
    Agent[AI Agent / Researcher] -- Natural Language Query --> Server[cBioPortal MCP Server]

    subgraph "MCP Tools Layer"
        Server --> T1[search_studies]
        Server --> T2[get_clinical_metadata]
        Server --> T3[fetch_molecular_profiles]
    end

    subgraph "cBioPortal Backend"
        T1 --> API[Web API / OpenAPI]
        T2 --> API
        T3 --> API
        API --> DB[(Genomic Database)]
    end

    subgraph "Candidate Groundwork"
        API -.-> PR12024[PR #12024 - Logic Fix]
        T2 -.-> PR12025[PR #12025 - Exception Guard]
        API -.-> PR12026[PR #12026 - Test Migration]
    end

Loading

Enabling Autonomous Genomic Discovery via MCP

I am Saurabh (saurabhhhcodes), and I'm proposing a self-defined GSoC project: Autonomous Genomic Agent Integration for cBioPortal via the Model Context Protocol (MCP).

Genomic data is inherently relational and complex. While the OpenAPI spec is great, AI agents still struggle with the high-dimensionality of the cBioPortal REST API. An MCP Server would provide a semantic bridge, mapping complex queries (e.g., "Find all Phase II Breast Cancer studies with high TP53 mutation counts") directly to optimized API tool calls.

To prepare for this, I've spent time modernizing the backend test suite and patching critical edge cases in the UUID service and Global Exception Handler (PRs #12024-#12027). I'd love to hear mentor thoughts on exposing the Portal's analytical power to agentic workflows!

CC: @jjgao @pieterlukasse @inmargivoni

[saurabhhhcodes]

Hey @jjgao and @pieterlukasse I look forward towards hearing any updates!?