Minimal MCP server implementation for performing meta-analyses using R packages (meta, metafor). Exposes endpoints for initializing sessions, uploading study data, performing meta-analysis, generating forest plots, assessing publication bias, and generating reports.
This MCP server has critical definition quality gaps across all dimensions. While tool names use reasonable verb_noun patterns (initialize_meta_analysis, upload_study_data, perform_meta_analysis, generate_forest_plot, assess_publication_bias, generate_report), the descriptions are extremely minimal and provide insufficient context for LLM-driven tool selection. Input schemas are visible in the TypeScript client but many parameters lack proper type constraints and descriptions. Most critically, output schemas are completely undocumented, there is no specification of what these tools actually return beyond placeholder responses in the R implementation. The R server code shows stub implementations returning minimal status objects with no real data structure or field documentation. Error handling is absent; there are no recovery guides or categorized error classifications. The implementation prioritizes minimal demo code over production-grade tool design.
Assess publication bias using specified methods
Generate forest plot from effect sizes and standard errors, or from stored analysis results
Generate analysis report in specified format
Initialize a new meta-analysis session
Perform meta-analysis on uploaded data
Upload and validate study data (CSV content in request body)
Output schemas completely missing for all 6 tools. LLMs cannot know what fields are returned, what IDs to extract for chaining, or how to interpret results.
Tool descriptions are too short and lack LLM-actionable context. Average description length is 42 chars; baseline for A-grade tools is 194 chars. Descriptions do not explain WHEN to call the tool or what it enables.
Parameter descriptions are minimal or missing. Examples: 'Data format (csv, excel, revman)' (28 chars) lacks explanation of what each format implies; 'Raw file content as string' is ambiguous (does it mean newline-delimited? UTF-8?). No validation constraints documented.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 41 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 48 | - | v1 |
No error handling or recovery guidance. R implementation shows basic validation (e.g., 'Both TE and seTE must be provided') but no mechanism for LLM to learn why a call failed or how to retry. No categorization of retryable vs user-fixable errors.
R implementation code contains placeholder stubs ('Placeholder returning the parameters. Actual analysis would load the data and call meta or metafor functions'). These are not production-ready; the server cannot actually perform meta-analyses.
Inconsistency between TypeScript client schema and R server implementation. Client expects tool parameters (e.g., TE, seTE arrays) but R implementation uses file-based state (tempfile uploads). LLM cannot predict which calling pattern is correct.
Missing enum constraints for categorical parameters. 'methods' in assess_publication_bias is an array of strings with no enum; 'plot_style' accepts free-form strings. LLMs will hallucinate invalid values.
No pagination or result limits documented. If meta-analysis returns 500 studies or generates a 50MB PDF, there's no guidance on chunking or streaming. Context window exhaustion risk.