Datacapt introduces MCP for AI-assisted study building and monitoring

[Lyon, France - 21 September 2026] Datacapt today released its Model Context Protocol (MCP) server. It gives an AI assistant 48 tools to build and test draft studies, then report on live ones, from plain-language requests. Data collected in live studies stays inside the Datacapt platform.

MCP is an open standard that connects AI applications to external systems. It gives an assistant the tools it needs to carry out a user’s request. Study teams connect an MCP-compatible assistant, such as Claude, and work under their own Datacapt permissions.

“Our customers want AI that works inside their study, under their own permissions. Datacapt MCP gives their assistant 48 tools to build a draft and report on a live study. Collected data stays in Datacapt.” Florentin Ory, CEO of Datacapt

1. Build and test a draft study

Users create draft studies, build eCRFs, add calculations and validation rules, prepare translations, and configure automations and randomization. A team could ask:

• “Add a BMI calculation to the screening form.”
• “Flag a systolic blood pressure above 180 mmHg on the vital signs form.”
• “Duplicate the Week 4 visit as Week 8.”
• “Add French and Spanish to the study and translate the forms.”
• “Set up 1:1 block randomization with two arms, stratified by site.”
• “Email the study coordinators each time a participant signs the eConsent.”
• “Create ten test participants and fill in the screening visit.”

With test participants, the team checks how each form behaves, then reviews the study and launches it in Datacapt. Teams can also save a form as a template or export the blank eCRF as a PDF.

2. Follow live studies

On a live study, the assistant reports on recruitment, open queries, missing data, source data verification (SDV), and ePRO and eConsent status, with detail by site or by record. A study manager could ask:

• “How many participants did each site include this month?”
• “Which sites have the most open queries?”
• “Which forms have missing data at the Lyon site?”
• “How much SDV is left, site by site?”
• “Which ePRO questionnaires are past their deadline?”
• “Which records need an investigator counter-signature on the consent?”

3. 48 tools under each user’s permissions

The server ships 48 tools. On the build side, they cover study setup, forms, translations, templates, automations, randomization and test participants. On the monitoring side, they cover recruitment, site performance, queries, missing data, SDV, ePRO, eConsent and audit reports.

The connection follows each user’s existing permissions. The assistant can change draft studies and nothing else. Datacapt records each change in the audit trail under that user’s name, with an MCP identifier.

4. Collected data stays in Datacapt

Data collected in a live study stays inside the Datacapt platform. The MCP server does not send participant answers to the AI assistant, so no AI model reads them, stores them or trains on them. The server also withholds participant identities and query text.

For monitoring, the assistant receives counts and statuses. The server raises any count below five to five, so a small number cannot point to one participant. The assistant cannot modify a live study, message participants or randomize them, and no tool reveals which arm a participant belongs to.

Availability

Datacapt has opened access to a limited number of customers. The Datacapt MCP guide explains how to set it up and use it.

To see Datacapt MCP in action on a study like yours, contact Datacapt to arrange a demo.

Background reading: EDC MCP servers: the four levels of AI agent access.

About Datacapt

Datacapt provides an eClinical platform for CROs, sponsors and research teams. Its solutions cover electronic data capture, patient-reported outcomes, eConsent, randomization, trial supply management and clinical operations, from study design through data collection and monitoring. Learn more at datacapt.com.