Owkin MCP Server

Provider-hosted MCP server that exposes Owkin's biology AI agents (HistoPLUS / Pathology Explorer) for H&E histopathology slide analysis across TCGA cohorts.

AI/ML by Owkin OAuth2 active

Connect (Remote Server) Docs

Overview

The Owkin MCP server connects Claude and other MCP-compatible clients to Owkin's K Pro platform, an agentic AI system for biology. The connector currently powers Pathology Explorer (built on the HistoPLUS model), an agent that transforms H&E pathology slides from the TCGA database into queryable insights for drug discovery, clinical research, and translational science.

Through the server, agents can detect and classify six cell types (lymphocytes, neutrophils, plasmocytes, fibroblasts, eosinophils, and cancer cells), compute quantitative histomics features such as cell counts, densities, nuclear morphology, regional areas, cell-to-cell co-occurrence, and TIL diffusivity, run cohort-level survival analyses, and retrieve slide thumbnails and tile-level predictions. The underlying models were trained on multimodal patient data sourced from Owkin's network spanning over 800 hospitals and 104 healthcare centers.

The integration was announced in January 2026 as part of Anthropic's Claude for Healthcare and Life Sciences launch. It is distributed both as a one-click Claude.ai connector and as a Claude Code plugin in the official anthropics/life-sciences marketplace. The MCP server itself is hosted by Owkin at https://mcp.k.owkin.com/mcp and authenticates users against their Owkin K Pro accounts.

Tools

Tool	Description
`describe_pathology_explorer`	Returns a high-level description of the Pathology Explorer agent and its capabilities.
`list_cell_types`	Lists the cell types that HistoPLUS can detect (lymphocytes, neutrophils, plasmocytes, fibroblasts, eosinophils, cancer cells).
`list_tcga_cohorts`	Lists the TCGA cancer cohorts available through K Pro (25 cohorts including breast, lung, bladder, colorectal carcinomas).
`query_slide_histomics_features`	Queries and filters slide-level histomics features (cell counts, densities, nuclear morphology, regional densities, cell-cell co-occurrence, tils_diffusivity).
`survival_analysis`	Runs cohort-level survival analysis to assess the prognostic value of histomics features.
`download_histomics_data`	Retrieves and downloads computed histomics data for further analysis.
`filter_tiles`	Filters tiles within a slide based on predictions or feature criteria.
`visualize_tile_predictions`	Visualizes tile-level predictions overlaid on pathology images.
`show_slide_thumbnail`	Displays the thumbnail image of a pathology slide.

Setup Guide

Prerequisites

An active Owkin K Pro account (sign up at https://www.owkin.com/)
A Claude.ai, Claude Desktop, or Claude Code workspace

Option 1: Claude.ai / Claude Desktop connector

For organization admins: Admin settings, then Connectors, then Browse, then Owkin, then Add to team.

For individual users: Settings, then Connectors, then find Owkin, then Connect, then sign in with your Owkin credentials.

If you are adding the connector manually, use:

Name: Owkin
URL: https://mcp.k.owkin.com/mcp

Option 2: Claude Code plugin

/plugin marketplace add anthropics/life-sciences
/plugin install owkin@life-sciences

Restart Claude Code and run /mcp to verify the Owkin server is connected.

Raw MCP config

The plugin definition in the anthropics/life-sciences repo is:

{
  "mcpServers": {
    "Owkin": {
      "type": "http",
      "url": "https://mcp.k.owkin.com/mcp"
    }
  }
}

Authentication is performed interactively through the connector flow against your Owkin K Pro account.

Use Cases

Quantify and compare lymphocyte, neutrophil, and cancer cell densities across TCGA breast or lung cohorts to find candidate response biomarkers.
Run cohort-level survival analysis stratified by a histomics feature (for example TIL diffusivity) to validate a prognostic hypothesis for drug discovery.
Characterize the spatial tumor microenvironment of a slide by analyzing cell-to-cell co-occurrence and regional densities.
Pull tile-level predictions and slide thumbnails into a Claude workflow to assemble figures for translational pathology reports.
Profile understudied cell populations such as eosinophils or plasmocytes across pan-cancer TCGA samples without leaving the Claude interface.

Example Prompts

"List the TCGA cohorts available in Pathology Explorer and tell me which ones include colorectal carcinoma."
"For TCGA-BRCA, compute lymphocyte density per slide and run a survival analysis stratifying patients by the median value."
"Show me the slide thumbnail and tile-level cancer cell predictions for the highest neutrophil-density slide in TCGA-LUAD."
"Filter slides in TCGA-BLCA where TIL diffusivity is in the top quartile and download the histomics features."
"Compare fibroblast and plasmocyte densities between two TCGA cohorts and summarize the differences."

Pros

Officially built and hosted by Owkin, with the MCP endpoint distributed through the Anthropic life-sciences marketplace.
Backed by HistoPLUS, a state-of-the-art pan-cancer cell detection model trained on data from over 800 hospitals.
Goes beyond text retrieval: exposes quantitative histomics, spatial analysis, survival statistics, and image visualization tools.
One-click install in Claude.ai or a single command in Claude Code.

Limitations

Requires a K Pro account from Owkin; the connector is not anonymous or self-hostable.
Data is currently limited to TCGA cohorts and the HistoPLUS feature set, not arbitrary user-uploaded slides.
A separate REST API spec, rate limits, and SDK documentation are not published; usage is intended through MCP clients.

Alternatives

10x Genomics MCP connector in the same Claude for Life Sciences marketplace for single-cell and spatial genomics data.
Benchling MCP for R&D notebook, registry, and assay data in life sciences workflows.
Self-hosted use of the open-source owkin/histoplus model directly, without the MCP layer, for teams that want to run cell detection on their own slides.