Introduction
Owkin build AI agents for biology to accelerate drug discovery and de-risk clinical trials. The Owkin connector currently powers HistoPLUS, an AI model that transforms H&E slides from the TCGA database into granular, queryable insights. Researchers can use it to quantify distinct cell types, analyze complex spatial tumor microenvironments, and validate hypotheses through cohort-level survival analysis. Use this connector to bridge the gap between raw biological data and actionable patient stratification, starting with histopathology.
You can connect to the Owkin MCP server through Claude.ai (web interface) or Claude Desktop (desktop application). Both options provide the same powerful capabilities - choose the one that fits your workflow.
Quick Setup
Prerequisites
An Owkin account with access to the K Pro platform
Access to Claude.ai or Claude Desktop
Account creation
Create a free account on Owkin K and use these credentials to sign in later. You can also use a K Pro account.
You can delay this by waiting for your client to authenticate to the remote connector.
Setup for Claude.ai
Configure the connector (admin)
Go to Admin settings → Connectors → Add custom connector
Enter the following details:
Name: Owkin
URL:
https://mcp.k.owkin.com/mcp
Do not fill any information into the advanced settings field (OAuth Cliend ID and Secret)
Add the custom connector
Connect (claude.ai or Claude Desktop)
Go to Settings → Connectors
Click on the Connect button associated to the Owkin connector
Allow Access to Owkin's MCP server
You will be redirected to an authentication page:
a. Enter your credentials and Sign in if you already created an account
b. Click on New user? Create an account otherwise
Connection should be enabled. Verify the connection by asking Claude in a new chat: "Can you list the available TCGA cohorts?"
If successful, Claude will retrieve and display the list of cohorts from the Owkin MCP server.
Troubleshooting Setup
Sign-in page hanging forever
While this is misleading, it shouldn't prevent the client to successfuly establish a connection to the MCP server. It will be fixed soon.
Error: Invalid or expired transaction
This error can appear after clicking twice on Allow Access on the Application Access Request form. This is a known issue with google chrome navigator due to the claude:// redirect uri. Using another web browser should fix the issue.
Connection fails or tools don't appear:
Verify you have an active Owkin account
Check that the MCP server URL is entered correctly
For Claude Desktop, ensure Node.js is installed on your system
Restart Claude after configuration changes
Authentication issues:
Contact your Owkin administrator to verify your access permissions
Ensure your K Pro platform credentials are active
Session expired:
Sessions currently last 15 minutes. If you see an error like "I'm encountering errors when trying to access the HistoPlus system right now", go to Settings → Connectors, disconnect and reconnect the Owkin connector. Automatic token refresh will be available shortly.
In any other case, please use this form to reach for support.
What is HistoPLUS?
Standard histology analysis overlooks the complex spatial organization of the tumor microenvironment (TME), which is critical for predicting treatment response.
Owkin's interoperable MCP transforms histology slides into queryable insights. Trained on 200k+ annotations, it segments and classifies all cells in minutes, enabling researchers to derive actionable, interpretable biomarkers.
Owkin's HistoPLUS MCP integration bridges this gap, transforming static images into queryable insights. This tool enables researchers to 'converse' directly with histological data to de-risk clinical trials. Powered by Owkin’s state-of-the-art HistoPLUS model (trained on 200k+ annotations, benchmarked against 6+ encoders and best-in-class architectures) it allows LLMs to reason over the TME with superior accuracy.
Key capabilities and example prompts
Refine patient stratification
Identify patient subgroups that generalist models miss through granular profiling of 6 distinct cell types (including understudied populations like neutrophils and eosinophils). Leverage spatial organization analysis to characterize TME structures and phenotypes beyond simple counts.
I'm looking for Lung Adenocarcinoma patients that might be resistant to immunotherapy. Are there cases with low immune infiltration in the TCGA cohort?Visualize whole-slide images
Build confidence in the model output by retrieving whole-slide images directly within the chat interface.
Find the slide the most enriched in eosinophils from cohort TCGA_BRCA and plot it.Assess prognostic value of H&E based markers
Test clinical hypotheses by performing survival analysis on your cohorts, by splitting patients based on features such as specific cell densities or spatial scores.
Is the density of plasmocytes associated with overall survival in bladder carcinoma?Extract quantitative evidence for reproducibility
Build trust in AI-generated insights by retrieving the underlying raw data for independent verification or downstream analysis.
Export the breakdown of all cell types for patient TCGA-A2-A0YI-01Z-00-DX1.1CF2EC2D-C722-467F-8832-409B823E8D8F.svs in parquet format, so I can reproduce this analysis.Understand HistoPLUS capabilities and context
Gain transparency into the model by querying its technical specifications directly. Learn about the supported cell types, the pan-cancer training dataset and more, to ensure the model is appropriate for your research question.
Can you provide an overview of the HistoPLUS model and its capabilities?
Components
Tools
HistoPlus description
Description: Provides an overview of the HistoPlus model and its capabilities.
Input: None
Output: Text description of the HistoPlus capabilities.
List available cell types
Description: Retrieves the list of cell types that the HistoPLUS model can detect in H&E slides.
Input: None
Output: List of cell types detected by the HistoPLUS model.
List TCGA cohorts
Description: Provides a list of The Cancer Genome Atlas (TCGA) cohorts available for analysis.
Input: None
Output: List of TCGA cohorts.
Describe slide-level histomics features
Description: Retrieves the description of slide-level histomics features.
Input: column_name (optional), cell_type (optional), name_only (optional, default False)
Output: If column_name is provided, return the description for that specific feature. If cell_type is provided, only return features related to that cell type. If neither is provided, return descriptions for all features. If name_only is True, return only the feature names, not the descriptions.
Filter slides by histomics features
Description: Filters slides based on specified histomics feature criteria.
Input: feature_name, cohort, descending (optional, default True), limit (optional, default 10)
Output: List of slide IDs that match the filtering criteria.
Get histomics features for a list of slides
Description: Retrieves histomics features for some slides.
Input: list of slide_id, list of column_name (optional)
Output: Histomics features associated with the specified slide ID. If column_name is provided, only return those features.
Display slide thumbnail
Description: Retrieves a thumbnail image of the specified slide.
Input: slide_id
Output: Thumbnail image of the slide.
Perform survival analysis
Description: Performs survival analysis based on specified histomics features.
Input: feature_name, cohort, endpoint (OS or PFS), stratify (optional, default False)
Output: Kaplan-Meier curves displaying survival analysis results.
Download histomics data
Description: Generate download link for histomics data for specified slides or cohort. If cohort is provided, all slides from the cohort will be included. If a list of slide_ids is provided, only those slides will be included. If neither are provided, all available TCGA slides will be included.
Input: list of slide_id (optional) or cohort (optional)
Output: presigned url to download the parquet file.
Last updated