mcp-server-redcap
Enables interaction with REDCap instances for managing research data, projects, and files through natural language.
README
mcp-server-redcap
A Model Context Protocol (MCP) server for interacting with REDCap (Research Electronic Data Capture) instances.
Exposes 38 tools covering data access, project design, file management, and longitudinal project structure. Designed for use with LLM agents that need to query or build REDCap projects in natural language conversations.
Tools
Records
| Tool | Description |
|---|---|
export_records |
Export records with optional filters by field, record ID, or event |
import_records |
Import records from a JSON payload |
delete_records |
Delete records by ID |
export_report |
Export a saved REDCap report by its ID |
export_field_names |
Export the mapping from field names to export names (required for checkbox fields) |
generate_next_record_name |
Get the next available record ID respecting the project's auto-numbering |
Metadata
| Tool | Description |
|---|---|
get_project_info |
Retrieve project-level information and settings |
get_metadata |
Export the data dictionary, optionally filtered by field or form |
export_logging |
Export the project audit log, with filters by type, user, record, or date range |
export_repeating_instruments_events |
Export the repeating instruments/events configuration |
import_repeating_instruments_events |
Update the repeating instruments/events configuration |
Instruments & project design
| Tool | Description |
|---|---|
get_instruments |
List all instruments with internal names and display labels |
get_instrument_event_mappings |
List arm/event/instrument mappings (longitudinal projects) |
add_field |
Add a new field to an instrument (round-trip through data dictionary) |
remove_field |
Remove a field from the data dictionary |
update_field |
Update specific properties of a field without touching the rest |
add_instrument |
Create a new instrument |
delete_instrument |
Delete an instrument and all its fields |
rename_instrument |
Rename an instrument's internal name |
move_field |
Reposition a field within the data dictionary |
clone_instrument |
Copy all fields of an instrument into a new one with a field-name prefix |
assign_instrument_to_event |
Add an instrument to a longitudinal event |
unassign_instrument_from_event |
Remove an instrument from a longitudinal event |
Analysis
| Tool | Description |
|---|---|
get_project_structure |
Compact overview of the whole project: instruments, field counts, event matrix |
check_field_references |
Find all fields that reference a given field in branching logic or calculations |
validate_branching_logic |
Scan all branching logic for references to non-existent fields |
Files
| Tool | Description |
|---|---|
export_file |
Download a file attachment from a record field (returned as base64) |
import_file |
Upload a file to a record field (accepts base64 content) |
delete_file |
Delete a file attachment from a record field |
export_pdf |
Export a PDF of one or more instruments for a record (returned as base64) |
Arms & events (longitudinal)
| Tool | Description |
|---|---|
export_arms |
List arms |
import_arms |
Create or update arms |
delete_arms |
Delete arms by number |
export_events |
List events, optionally filtered by arm |
import_events |
Create or update events |
delete_events |
Delete events by unique name |
Surveys
| Tool | Description |
|---|---|
export_survey_link |
Get a participant-specific survey URL for a record |
export_survey_participant_list |
List survey participants and their response status |
Installation
pip install mcp-server-redcap
Or with uv:
uvx mcp-server-redcap
REDCap version compatibility
The server works against any REDCap instance running version 8.0 or later. Most tools are available from version 6.x, but export_pdf requires 8.x. Repeating instruments (export_repeating_instruments_events, import_repeating_instruments_events) require 6.16+. Tools that are unavailable on a given instance return a plain error string rather than crashing.
The current REDCap version of the connected instance is included in the output of get_project_structure.
Configuration
The server reads connection details from environment variables:
| Variable | Required | Description |
|---|---|---|
REDCAP_URL |
Yes | Full REDCap API endpoint URL (e.g. https://redcap.example.org/api/) |
REDCAP_TOKEN |
Yes | Project-level API token from REDCap → API → Generate Token |
REDCAP_VERIFY_SSL |
No | Set to false to skip SSL verification (default: true) |
Create a .env file in your working directory and the server will load it automatically:
REDCAP_URL=https://redcap.example.org/api/
REDCAP_TOKEN=your_project_token_here
Usage with Claude Code
Add the server to your Claude Code MCP configuration:
claude mcp add redcap -e REDCAP_URL=https://redcap.example.org/api/ \
-e REDCAP_TOKEN=your_token -- uvx mcp-server-redcap
Or edit .claude/settings.json manually:
{
"mcpServers": {
"redcap": {
"command": "uvx",
"args": ["mcp-server-redcap"],
"env": {
"REDCAP_URL": "https://redcap.example.org/api/",
"REDCAP_TOKEN": "your_project_token_here"
}
}
}
}
Development
git clone https://github.com/msicilia/mcp-server-redcap
cd mcp-server-redcap
uv sync --extra dev
Run the MCP inspector for interactive local testing:
uv run mcp dev src/mcp_server_redcap/server.py
Run the server directly (stdio transport, for use with MCP clients):
uv run mcp-server-redcap
Project structure
src/mcp_server_redcap/
├── __init__.py
├── __main__.py # python -m mcp_server_redcap
├── server.py # FastMCP server factory and entry point
├── connection.py # REDCap project connection (lazy, singleton)
└── tools/
├── __init__.py
├── records.py # record CRUD, field names, next record name
├── metadata.py # project info, data dictionary, logging, repeating instruments
├── instruments.py # instrument listing, field and instrument mutations
├── analysis.py # project structure overview, reference and logic validation
├── files.py # file attachments and PDF export
├── arms_events.py # longitudinal arms and events CRUD
└── surveys.py # survey links and participant lists
See DESIGN.md for the full API reference and design rationale.
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。