ClinicalTrials.gov MCP Server

An MCP (Model Context Protocol) server that exposes the ClinicalTrials.gov v2 API as tools for LLM agents. Supports searching, retrieving, and analyzing clinical study data from the full registry of 570,000+ studies.

Tools

clinicaltrials_get_study

Retrieve a single study record by NCT ID.

Parameter	Type	Required	Description
nct_id	string	Yes	NCT identifier, e.g. NCT00000102
fields	StudyField[]	No	Specific fields to return; omit for all fields
format	json | csv	No	Response format (default: json)
markup_format	markdown | legacy	No	Text markup format (default: markdown)

clinicaltrials_search_studies

Search studies by condition, intervention, sponsor, location, and more. Returns paginated results.

Query parameters (at least one required):

Parameter	Description
query_cond	Condition or disease (e.g. diabetes)
query_term	General keyword search
query_intr	Intervention or treatment name
query_titles	Search within study titles only
query_id	NCT ID or other study identifier
query_spons	Sponsor or collaborator name
query_locn	Location name or city
query_patient	Plain-language patient-friendly search

Filters:

Parameter	Description
filter_overall_status	List of recruitment statuses (e.g. [RECRUITING, NOT_YET_RECRUITING])
filter_geo	Geographic radius, e.g. distance(39.0,-77.0,50mi)
filter_ids	List of specific NCT IDs
post_filter_overall_status	Status filter applied after aggregation counts
post_filter_geo	Geo filter applied after aggregation counts
agg_filters	Aggregation filters string, e.g. phase:2 3,studyType:int

Pagination & output:

Parameter	Default	Description
sort	—	Sort field and direction, e.g. LastUpdatePostDate:desc
page_size	20	Results per page (1–1000)
page_token	—	Cursor from previous response's nextPageToken
count_total	false	Include total match count in response
fields	all	Specific StudyField values to return
format	json	json or csv
markup_format	markdown	markdown or legacy

clinicaltrials_get_field_values

Get the distribution of values across all studies for one or more fields. Useful for understanding what values exist and how common they are (e.g. study counts per phase, per country, per status).

Parameter	Type	Required	Description
fields	StudyField[]	Yes	Fields to get distributions for

Works best with enumerable fields: Phase, OverallStatus, StudyType, Sex, StdAge, LeadSponsorClass, InterventionType, LocationCountry, DesignAllocation, IsFDARegulatedDrug, HasResults, IPDSharing.

Field Selection

All three tools accept a fields parameter typed as List[StudyField]. The StudyField enum contains all 342 valid field names sourced from the ClinicalTrials.gov metadata API, grouped by section:

• Identification: NCTId, BriefTitle, OfficialTitle, Acronym, OrgFullName, ...
• Status: OverallStatus, StartDate, CompletionDate, LastUpdatePostDate, ...
• Sponsor: LeadSponsorName, LeadSponsorClass, CollaboratorName, ...
• Design: Phase, StudyType, DesignAllocation, DesignMasking, EnrollmentCount, ...
• Arms / Interventions: InterventionType, InterventionName, ArmGroupLabel, ...
• Outcomes: PrimaryOutcomeMeasure, SecondaryOutcomeMeasure, ...
• Eligibility: EligibilityCriteria, Sex, MinimumAge, MaximumAge, StdAge, ...
• Locations: LocationFacility, LocationCity, LocationState, LocationCountry, ...
• Results: participant flow, baseline characteristics, outcome measures, adverse events
• MeSH / Browse: ConditionMeshTerm, InterventionMeshTerm, ...

Setup

Prerequisites

• Python 3.13+
• uv

Install

git clone https://github.com/your-org/nih-clinicaltrials-mcp-server
cd nih-clinicaltrials-mcp-server
uv sync
uv pip install -e .

Run locally (stdio)

uv run python src/clinicaltrials/app.py

Run as HTTP server

PORT=8000 uv run python src/clinicaltrials/app.py
# or
uv run uvicorn clinicaltrials.app:app --port 8000

Connect from Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "clinicaltrials": {
      "command": "uv",
      "args": [
        "run",
        "--directory", "/path/to/nih-clinicaltrials-mcp-server",
        "python", "src/clinicaltrials/app.py"
      ]
    }
  }
}

Project Structure

src/clinicaltrials/
├── app.py       # FastMCP server init and transport config
├── tools.py     # Tool implementations (register_tools)
├── models.py    # Pydantic models and enums (StudyField, OverallStatus, ...)
├── utils.py     # Shared HTTP client (Chrome TLS impersonation) and error handling
├── prompts.py   # MCP prompts
└── routes.py    # Custom HTTP routes (/health)

Implementation Notes

TLS fingerprinting: The ClinicalTrials.gov CDN blocks standard Python HTTP clients via JA3 TLS fingerprint detection. This server uses curl-cffi to impersonate Chrome's TLS handshake, which resolves the 403 errors that httpx and requests produce.

No authentication required: The ClinicalTrials.gov API is public and requires no API key.

API version: ClinicalTrials.gov v2 (https://clinicaltrials.gov/api/v2). Data is refreshed daily.