Skip to main content
Glama

nonmem-mcp-server

MCP (Model Context Protocol) server for NONMEM pharmacometric modeling workflows. Gives Claude (and any MCP-compatible client) structured access to NONMEM models, results, and simulation tools.

Supports Windows, macOS, and Linux.

Features

Parsing & Analysis (no NONMEM needed)

  • read_ext_file — Parse .ext files for parameter estimates, SEs, OFV, condition number

  • read_lst_file — Extract termination status, shrinkage, covariance step results

  • parse_control_stream — Structural parsing of .ctl/.mod files (THETAs, OMEGAs, $EST options)

  • read_nm_dataset — Dataset summary: subjects, observations, missing values

  • read_nm_tables — Parse SDTAB/PATAB with statistics for CWRES, ETAs, PRED

  • compare_models — Multi-run OFV comparison with delta-OFV and AIC

  • summarize_run — Combined .ctl + .ext + .lst summary

  • list_runs — Scan project directories for NONMEM runs

Execution & Diagnostics

  • submit_run — Start NONMEM runs (async, fire-and-poll pattern)

  • check_run_status — Monitor iteration progress via .ext file

  • get_run_results — Retrieve parsed results when complete

  • cancel_run — Kill running NONMEM jobs

  • run_diagnostics — Automated checks: boundary, condition number, shrinkage, RSE

  • execute_psn_vpc — Run VPC via PsN (predcorr, stratify, lloq options)

  • execute_psn_bootstrap — Run bootstrap via PsN (BCa, stratify)

  • check_psn_status — Monitor PsN job progress

  • parse_psn_results — Parse existing PsN output directories (no installation needed)

  • check_nonmem_setup — Detect NONMEM, PsN, R installation status

Simulation (no NONMEM needed)

  • translate_to_mrgsolve — Convert NONMEM .ctl/.mod to mrgsolve model code

  • simulate_mrgsolve — Run PK simulations via mrgsolve (R)

  • generate_vpc_data — Generate VPC data using mrgsolve + vpc R package

  • check_r_setup — Check R and package availability

Prompts

  • review_model — Model review checklist

  • interpret_results — Pharmacological interpretation

  • troubleshoot_run — Diagnose run failures

  • suggest_next_model — Suggest next modeling steps

  • write_methods_section — Draft publication Methods text

Related MCP server: simple-fal-mcp

Requirements

  • Python 3.12+

  • uv (recommended) or pip

Optional

  • NONMEM 7.3–7.6 — Required for submit_run (commercial license)

  • PsN — Required for execute_psn_vpc, execute_psn_bootstrap

  • R with mrgsolve, vpc, dplyr, ggplot2 — Required for simulation and GOF tools

Installation

git clone https://github.com/sueinchoi/nonmem-mcp-server.git
cd nonmem-mcp-server
uv sync

Using pip

pip install git+https://github.com/sueinchoi/nonmem-mcp-server.git

Or for development:

git clone https://github.com/sueinchoi/nonmem-mcp-server.git
cd nonmem-mcp-server
pip install -e .

NONMEM Path Configuration

The server auto-detects NONMEM from common install locations. If auto-detection fails, set one of these environment variables:

Variable

Description

Example

NONMEM_NMFE_PATH

Full path to nmfe executable

/opt/nm760/run/nmfe76

NONMEM_INSTALL_PATH

NONMEM installation root

/opt/nm760

Auto-detected paths

macOS / Linux:

  • /opt/nm760/run/nmfe76, /opt/NONMEM/nm75/run/nmfe75, etc.

  • /usr/local/NONMEM/nm76/run/nmfe76

  • ~/NONMEM/nm76/run/nmfe76

Windows:

  • C:\nm760\run\nmfe76.bat, C:\NONMEM\nm76\run\nmfe76.bat

  • C:\Program Files\NONMEM\nm76\run\nmfe76.bat

  • D:\NONMEM\nm76\run\nmfe76.bat

Usage with Claude Code

Basic (no NONMEM)

claude mcp add nonmem -- nonmem-mcp

With uv (from source)

claude mcp add -s user nonmem -- \
  uv run --directory /path/to/nonmem-mcp-server python -m nonmem_mcp

With NONMEM path

# macOS / Linux
claude mcp add -s user \
  -e NONMEM_NMFE_PATH=/opt/nm760/run/nmfe76 \
  nonmem -- \
  uv run --directory /path/to/nonmem-mcp-server python -m nonmem_mcp
# Windows (PowerShell)
claude mcp add -s user `
  -e NONMEM_NMFE_PATH=C:\nm760\run\nmfe76.bat `
  nonmem -- `
  uv run --directory C:\path\to\nonmem-mcp-server python -m nonmem_mcp

Verify

claude mcp list
# nonmem: ... - ✓ Connected

Usage with Claude Desktop

Add to claude_desktop_config.json:

macOS / Linux

{
  "mcpServers": {
    "nonmem": {
      "command": "uv",
      "args": ["run", "--directory", "/path/to/nonmem-mcp-server", "python", "-m", "nonmem_mcp"],
      "env": {
        "NONMEM_NMFE_PATH": "/opt/nm760/run/nmfe76"
      }
    }
  }
}

Windows

{
  "mcpServers": {
    "nonmem": {
      "command": "uv",
      "args": ["run", "--directory", "C:\\path\\to\\nonmem-mcp-server", "python", "-m", "nonmem_mcp"],
      "env": {
        "NONMEM_NMFE_PATH": "C:\\nm760\\run\\nmfe76.bat"
      }
    }
  }
}

Examples

# Summarize a NONMEM run
"Summarize the run in /path/to/run001/"

# Compare covariate models
"Compare OFV across all models in the covariate analysis directory"

# Diagnose a failed run
"Why did this run fail? Check /path/to/run.lst"

# Systematic model development
"Develop a 2-compartment model with forward IIV addition"

# Translate to mrgsolve for simulation
"Convert my NONMEM model to mrgsolve and run a VPC"

# GOF plots
"Generate GOF plots for run015 with IPRED and CWRES"

Capability Matrix

Feature

No NONMEM

+ NONMEM

+ PsN

Parse .ext/.lst/.ctl

✓

✓

✓

Model comparison

✓

✓

✓

Diagnostics

✓

✓

✓

mrgsolve simulation

✓

✓

✓

mrgsolve VPC

✓

✓

✓

NONMEM execution

✗

✓

✓

PsN VPC

✗

✗

✓

PsN Bootstrap

✗

✗

✓

Parse PsN results

✓

✓

✓

Included Examples

The examples/theopp/ directory contains a complete model development workflow using the Theophylline dataset:

  • run001.ctl — 1-compartment base model (ADVAN2, FO)

  • run002.ctl — 2-compartment model (ADVAN4 TRANS4)

  • run003–005.ctl — 2-comp base models with different initial estimates (no IIV)

  • run006–010.ctl — Single IIV forward addition (CL, V2, Q, V3, KA)

  • run011–014.ctl — Double IIV forward addition (V2+CL, V2+KA, V2+Q, V2+V3)

  • run015.ctl — Final 1-comp model with FOCE+INTER, IPRED, CWRES

  • gof_plot_v2.R — GOF plotting script (DV vs PRED/IPRED, CWRES, QQ plot)

License

MIT

Available Tools

22 tools
cancel_runA

Cancel a running NONMEM job.

Inputs: job_id (required).

check_nonmem_setupA

Check NONMEM, PsN, R, and mrgsolve installation status and paths.

Inputs: no named parameters.

check_psn_statusC

Check status and progress of a PsN VPC or bootstrap job.

Inputs: job_id (required).

check_r_setupA

Check R installation and required packages (mrgsolve, vpc, dplyr, ggplot2).

Inputs: no named parameters.

check_run_statusB

Check the status and progress of a submitted NONMEM run by job ID.

Inputs: job_id (required).

compare_modelsA

Compare multiple NONMEM runs by OFV, delta-OFV, parameter counts, and AIC. Accepts a list of run directories or .ext file paths.

Inputs: models (required).

execute_psn_bootstrapA

Run bootstrap using PsN (async). Returns job ID for polling. Requires PsN + NONMEM.

Inputs: model_path (required), samples, options.

execute_psn_vpcA

Run VPC using PsN (async). Returns job ID for polling. Requires PsN + NONMEM.

Inputs: model_path (required), samples, options.

generate_vpc_dataA

Generate VPC data using mrgsolve + vpc R package. No NONMEM needed. Requires observed data and model.

Inputs: observed_data_path (required), model_code, model_file, n_sim, +4 more.

get_run_resultsB

Get full parsed results of a completed NONMEM run by job ID.

Inputs: job_id (required).

list_runsA

Scan a project directory for NONMEM run files and list their status (completed/failed/running).

Inputs: project_dir (required), recursive.

parse_control_streamA

Parse a NONMEM control stream (.ctl/.mod) into structured components: $THETA, $OMEGA, $SIGMA, $EST options, input columns, data file, and ADVAN/TRANS.

Inputs: file_path (required).

parse_psn_resultsA

Parse results from an existing PsN output directory (VPC or bootstrap). No PsN/NONMEM installation needed.

Inputs: results_dir (required).

read_ext_fileA

Parse a NONMEM .ext file to extract parameter estimates, standard errors, OFV, eigenvalues, and condition number.

Inputs: file_path (required).

read_lst_fileA

Parse a NONMEM .lst file for termination status, minimization success, shrinkage, condition number, and warnings.

Inputs: file_path (required).

read_nm_datasetA

Read and summarize a NONMEM dataset: subject count, observation count, dose records, missing values, and column overview.

Inputs: file_path (required), input_columns.

read_nm_tablesB

Parse NONMEM table output files (SDTAB, PATAB, etc.) and compute summary statistics for key columns (CWRES, PRED, IPRED, ETAs).

Inputs: file_path (required).

run_diagnosticsA

Run comprehensive post-run diagnostics: boundary checks, condition number, shrinkage, precision (RSE), convergence warnings. No NONMEM needed.

Inputs: run_dir (required), run_name.

simulate_mrgsolveA

Run a simulation using mrgsolve (R). No NONMEM needed. Provide model code or model file path.

Inputs: model_code, model_file, data_path, n_subjects, +6 more.

submit_runA

Submit a NONMEM run (async). Returns a job ID for polling. Requires NONMEM installation.

Inputs: ctl_path (required), work_dir, nmfe_path, run_name.

summarize_runB

Generate a comprehensive summary of a NONMEM run by combining control stream, .ext, and .lst results.

Inputs: run_dir (required), run_name.

translate_to_mrgsolveA

Translate a NONMEM control stream (.ctl/.mod) to mrgsolve model code. Handles ADVAN1-13, THETAs, OMEGAs, $PK, $DES, $ERROR.

Inputs: ctl_path (required), output_path, parameter_values.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.6/5.0

    Scored across 22 tools

    Disambiguation4/5

    Most tools have clearly distinct purposes, such as 'cancel_run' for job cancellation versus 'submit_run' for job submission. However, some overlap exists, like 'check_run_status' and 'list_runs' both providing run status information, which could cause mild confusion in selection.

    Naming Consistency5/5

    Tool names follow a highly consistent verb_noun pattern throughout, such as 'cancel_run', 'check_nonmem_setup', and 'parse_control_stream'. There are no deviations in naming conventions, making the set predictable and readable.

    Tool Count3/5

    With 22 tools, the count is borderline high for a NONMEM-focused server, potentially feeling heavy. While many tools are specialized for tasks like diagnostics and parsing, the number might overwhelm agents, though it aligns with the comprehensive scope of NONMEM modeling workflows.

    Completeness5/5

    The tool surface provides complete coverage for NONMEM modeling, including setup checks, job submission, status monitoring, result parsing, diagnostics, and simulation. There are no obvious gaps, supporting full lifecycle management from data input to analysis and visualization.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      C
      maintenance
      An MCP server that gives LLMs native access to cheminformatics and molecular ML tools, enabling molecular structure manipulation, descriptor calculation, ML model training, and analysis report generation through natural conversation.
      37
      MIT
    • A
      license
      Not graded
      quality
      D
      maintenance
      An MCP server that provides tools for reproducible COMSOL Multiphysics batch workflows, enabling AI agents to compile, run, and diagnose Java API model scripts with COMSOL.
      5
      MIT