Robins I MCP server
{
"mcpServers": {
"blackswancausallabs-robins-i": {
"command": "uvx",
"args": [
"robins-i-mcp"
]
}
}
}{
"mcpServers": {
"blackswancausallabs-robins-i": {
"command": "uvx",
"args": [
"robins-i-mcp"
]
}
}
}{
"mcpServers": {
"blackswancausallabs-robins-i": {
"command": "uvx",
"args": [
"robins-i-mcp"
]
}
}
}{
"servers": {
"blackswancausallabs-robins-i": {
"command": "uvx",
"args": [
"robins-i-mcp"
]
}
}
}{
"mcpServers": {
"blackswancausallabs-robins-i": {
"command": "uvx",
"args": [
"robins-i-mcp"
]
}
}
}file .mcp.json (in your project root)file claude_desktop_config.jsonfile ~/.cursor/mcp.jsonfile .vscode/mcp.jsonfile ~/.codeium/windsurf/mcp_config.jsoninstall uvx robins-i-mcpversion 0.1.0
What it does
An MCP server that implements ROBINS-I V2, a structured assessment tool for evaluating risk of bias in non-randomized cohort studies with follow-up structure. It enforces deterministic, auditable judgements by matching evidence to text offsets, searching for absence rather than accepting claims, and computing domain scores from explicit algorithms.
- Assess risk of bias in a target-trial emulation study by ingesting its PDF and supplements, answering signalling questions grounded in the text, and generating a stamped report.
- Conduct a systematic review of N cohort studies by running multiple assessments and exporting all records to a robvis visualization.
- Override a computed domain judgement with human expertise while preserving both the algorithmic result and the reviewer's justification in the report.
Tools this server exposes
| Tool | Does |
|---|---|
| get_spec | Returns the ROBINS-I V2 specification with optional compact or full detail. |
| parse_document | Ingests a PDF, docx, or text document plus supplements and returns a hash with a cue survey of where evidence for each domain may appear. |
| parse_pmcid | Retrieves a study from Europe PMC by PMCID and prepares it for assessment. |
| set_prespecified_confounders | Sets the reviewer's list of important confounding factors required to score domain 1; blocks assessment until this gate is passed. |
| specify_result | Declares the target result and selects whether domain 1 assesses intention-to-treat (variant A) or per-protocol (variant B) analysis. |
| assess_result | Presents the scaffold of signalling questions for a given domain, accepting answers grounded in text quotations with character-offset verification. |
| submit_answers | Submits answers for a domain; when domain=0, finalizes the assessment, computes all domain and overall judgements, and emits a portable record. |
| render_report | Re-renders the stamped assessment report from a saved record. |
| export_robvis | Aggregates records from multiple assessment runs and exports them as a CSV formatted for robvis visualization. |
Before you enable it
An MCP server runs with whatever access you give it. Start with a development account rather than production, and check which tools the server exposes before you let an agent call them unattended.