MCP Server for GROMACS 2025.4 molecular dynamics simulations. Provides both synchronous and asynchronous (submit) APIs for GROMACS tools including MD simulations, trajectory analysis, TPR file analysis, and workflow management.
This GROMACS MCP server has moderate definition quality with significant gaps. Tools are explicitly registered with @mcp.tool() decorators and have descriptions, but parameter descriptions lack specificity, error handling is minimal, and there are no output schemas documented. The server handles domain-specific operations (MD simulations, GROMACS analysis) but the tool interface is not optimized for LLM reasoning. Naming is generally clear (verb_noun), but parameter constraints are under-specified and error responses lack recovery guidance. This server would benefit from enhanced parameter documentation, explicit output schemas, and actionable error messages.
Analyze GROMACS TPR files and extract simulation parameters. Fast operation that completes in seconds. Use for examining TPR file contents, simulation parameters, and system information.
Cancel a running job.
Get log output from a running or completed job.
Get the results of a completed job.
Get the status of a submitted job.
List all submitted jobs.
Execute GROMACS command-line tools with flexible parameter handling. Fast operation for running GROMACS analysis commands like rms, rmsf, energy. Supports all common GROMACS commands with automatic parameter handling.
Output schemas not documented. No tool returns a documented structure, LLMs cannot plan downstream tool calls or know what fields to extract. Only `dict` return types are shown without field specifications.
Parameter descriptions lack specificity. For example, 'tail' in get_job_log says 'Number of lines from end (default: 50, use 0 for all)' but does not specify minimum/maximum bounds or whether negative values are rejected. 'input_files' and 'output_files' are described as 'Dict of input/output files with flags' but do not specify the exact dictionary key/value format or allowed flags.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 53 | <=2025-11-25 | v2 |
| 2026-03-09 | D | 56 | - | v1 |
Create and manage GROMACS simulation workflows. Fast operation for running predefined workflows like 'simple_md', 'analysis', or 'demo'. Manages workflow dependencies and execution order automatically.
Submit batch analysis for multiple input files. Processes multiple TPR files or trajectories in a single background job. Suitable for analyzing many simulation files at once.
Submit molecular dynamics simulation for background processing. This operation can take significant time (minutes to hours) depending on simulation length. Returns a job_id for tracking progress.
Error handling is minimal. The code returns format_tool_result(False, error_message=...) but does not categorize errors as retryable vs. fatal, nor does it provide recovery guidance (e.g., 'Try submit_md_simulation with fewer steps' or 'Check that the TPR file is valid with analyze_tpr first').
No enum constraints on parameter values. 'workflow_type' accepts 'simple_md', 'analysis', 'demo' and 'analysis_type' accepts 'tpr', 'trajectory', but these are described as free-form strings, not enums. Enums would prevent LLM hallucination of invalid values.
No pagination support documented. 'list_jobs' returns a list with an optional status filter, but no limit, offset, or cursor parameters. If the job list grows large, the response could explode the context window. The tool description does not mention a result cap.
Missing permission gates and audit trails. Destructive tools like 'cancel_job' and 'submit_md_simulation' (which consumes compute) do not declare required permissions (e.g., 'write:job', 'execute:simulation'). No audit logging is visible in the code.
Input validation is deferred to called modules. The code checks for FileNotFoundError and ValueError, but validation logic is in external functions (validate_input_file, setup_output_directory). If validation fails, error messages are generic ('File not found: [error]'). No inline constraints on file paths (e.g., disallow '../' for path traversal).
No dry-run or confirmation step for irreversible operations. 'submit_md_simulation' and 'cancel_job' modify state but offer no confirm_before_execute pattern. An agent mistake could cancel a multi-hour simulation.