Psychometrics MCP
Psychometrics MCP is a local-first, measurement-aware MCP server for reproducible statistical and psychometric analysis with constrained, non-arbitrary tools.
Check whether local Python and the fixed R/eRm Rasch engine are available.
Audit response data for dimensions, missingness, categories, ranges, and zero-variance items.
Compute classical test theory item analysis: item summaries, item-rest correlations, raw alpha, and SEM.
Generate descriptive statistics with sample flow, missingness, quartiles, and robust boundaries.
Compute Pearson or Spearman correlation matrices with pairwise/listwise deletion and pair-specific n.
Fit numeric ordinary least squares regression with classical inference, fit, sample flow, and influence diagnostics.
Run confirmatory factor analysis with fixed lavaan CFA ML/MLR estimation.
Perform seeded Horn-style parallel analysis for factor retention evidence.
Fit exploratory factor analysis with fixed MINRES/ML extraction and oblimin/varimax/no rotation.
Build purpose- and design-aware psychometric analysis plans.
Fit a fixed dichotomous Rasch model using eRm::RM with CML estimation.
All tools use strict input schemas and fixed R scripts; no arbitrary R or shell code is accepted.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Psychometrics MCP Run item analysis and a Rasch model on my response data, then summarize the results."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Psychometrics MCP
Psychometrics MCP is a local-first, measurement-aware MCP server for reproducible statistical and psychometric analysis. It is designed to help an AI host use constrained analytical tools without exposing arbitrary R or shell execution.
The project is an early prototype. Current results require qualified human review and must not be used as automatic evidence of score validity, fairness, or fitness for consequential decisions.
Current tools
Tool | Purpose |
| Check Python and fixed R engine availability |
| Audit dimensions, missingness, categories, ranges, and variance |
| Item summaries, item-rest correlations, raw alpha, and SEM |
| Available-case numeric summaries, quartiles, and missingness |
| Pearson or Spearman matrices with pairwise/listwise deletion and pair-specific n |
| Unsmoothed two-step polychoric matrix for ordered categorical variables |
| Numeric OLS with classical inference, fit, sample flow, and influence diagnostics |
| Continuous-indicator simple-structure CFA with fixed |
| Ordered-indicator simple-structure CFA with fixed |
| Configural, metric, scalar, and strict multi-group CFA comparisons without automatic pass/fail rules |
| Seeded Horn-style common-factor retention evidence with MINRES or ML |
| Seeded permutation PA with polychoric/Pearson, PCA/common-factor, and cutoff sensitivity |
| Fixed-factor continuous EFA with MINRES/ML and oblimin/varimax/no rotation |
| Fixed-factor EFA from an unsmoothed polychoric matrix |
| Build a purpose- and design-aware analysis sequence |
| Fit a fixed dichotomous |
Rasch is the first end-to-end numerical validation slice, not the product boundary. The development plan includes descriptive statistics, correlations, regression, measurement-aware machine learning, factor models, broader IRT/Rasch, DIF and invariance, linking/equating, CAT/MST, G-theory, rater models, DCM, validity evidence, and reproducible reporting.
Related MCP server: R MCP Server
Local installation
Requirements:
Python 3.11+
R with
eRm,lavaan,psych,GPArotation, andjsonlitefor the fixed Rasch, CFA, parallel-analysis, and EFA adapters
python3 -m venv .venv
source .venv/bin/activate
python -m pip install '.[dev]'
psychometrics-mcpThe command starts a stdio MCP server. Configure an MCP host to launch the
absolute path to .venv/bin/psychometrics-mcp.
Docker
docker build -t psychometrics-mcp:local .
docker run --rm -i psychometrics-mcp:localThe server communicates over stdin/stdout, so -i is required. It runs as a
non-root user and does not need a data volume because tools accept structured
inputs. If future file-based tools are enabled, mount only a narrow directory
read-only.
Verification
ruff check .
pytest
pip-audit
python -m buildAll tool results carry schema_version: "1.0". The test suite includes real
eRm, lavaan, and psych integration tests and versioned numerical
regression targets.
Each reference states whether it is independent or engine-based validation. CI
separately validates Python behavior, the R adapters, and the Docker build.
Interpretation and safety boundaries
Tool inputs are strict schemas; unknown fields are rejected.
R-backed tools call fixed scripts and never accept user R or shell code.
Missingness, exclusions, methods, versions, warnings, and sample flow are returned with analytical results.
Reliability is not validity, model fit is not fairness, and prediction is not measurement of a construct.
Local deployment is preferred for item-response and student-level data.
See Architecture, Tool Admission Policy, OLS Regression, Confirmatory Factor Analysis, Measurement Invariance, EFA and Parallel Analysis, Ordinal Correlation and Factor Models, Privacy and Deployment, the Output Contract, and the Development Backlog.
License
MIT
Available Tools
7 toolscheck_computation_capabilitiesA
Report whether local Python and the fixed R/eRm Rasch engine are available.
Inputs: no named parameters. Returns: object
correlation_matrixA
Compute Pearson or Spearman correlations with explicit missing-data handling.
Inputs: request (required). Returns: object
ctt_item_analysisC
Compute item summaries, item-rest correlations, raw alpha, and SEM with warnings.
Inputs: data (required). Returns: object
descriptive_statisticsB
Summarize numeric variables with sample flow, missingness, and robust boundaries.
Inputs: data (required). Returns: object
inspect_response_dataB
Audit response shape, missingness, categories, ranges, and zero-variance items.
Inputs: data (required). Returns: object
plan_psychometric_analysisC
Create a measurement-aware analysis sequence from the intended use and design.
Inputs: request (required). Returns: object
rasch_modelA
Fit a fixed dichotomous Rasch model with eRm::RM; arbitrary code is never accepted.
Inputs: data (required). Returns: object
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 7 tools
The core analysis tools are mostly distinct, but inspect_response_data, descriptive_statistics, and ctt_item_analysis all touch on summarizing data, missingness, and item-level properties. Descriptions help clarify intent, yet an agent could still hesitate between an audit and a general descriptive summary.
Tool names use snake_case throughout and are readable, but they mix verb-led names like inspect_response_data and plan_psychometric_analysis with noun-led names like correlation_matrix and rasch_model. The inconsistency is mild but prevents a clear predictable verb_noun convention.
Seven tools is well-scoped for a psychometrics-focused server covering environment checks, data inspection, descriptive stats, correlations, CTT, planning, and Rasch modeling. Each tool earns its place without bloating the surface.
The server covers a coherent CTT-plus-Rasch workflow from planning through data auditing, descriptives, correlations, item analysis, and model fitting. Minor gaps exist such as no polytomous IRT or visualization tools, but the core measurement workflow is not blocked.
Maintenance
Related MCP Connectors
Find, vet, and run MCP tools through a secure audited gateway with prompt-injection risk scoring
MCP-native AI evaluation: rubric audits, eval suites, and proof reports for AI/LLM output.
Bounded tools for rendering, extraction, RAG, enrichment, local discovery and review analysis.
Deterministic validation for AI-generated artifacts: JSON Schema, OpenAPI response, SQL syntax.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceEnables AI agents to perform statistical process control calculations using validated, deterministic tools such as control charts, capability analysis, and tolerance intervals.3MIT
- AlicenseNot gradedqualityDmaintenanceEnables AI assistants to execute R code, create visualizations, analyze data, and manage packages through a local Rscript CLI.MIT
- AlicenseAqualityCmaintenanceEnables AI agents to perform reproducible, verifiable statistical analysis through 25 deterministic tools for descriptive statistics, hypothesis testing, regression, clustering, time-series forecasting, and Chinese-labeled plotting.301MIT
- FlicenseNot gradedqualityCmaintenanceEnables secure execution of statistical Python code in isolated micro-VM sandboxes, generating charts and analytical summaries.-