Model Context Protocol (MCP) Server for Amber MD Simulations. Provides tools for running molecular dynamics simulations using the Amber software suite.
Three tools with consistent naming patterns (verb_noun: amber_run_*, amber_prepare_*, amber_generate_*). All tools have substantive descriptions (120 - 180 chars, within the 34 - 392 baseline). Input schemas are fully defined with proper JSON Schema types and parameter descriptions. However, OUTPUT schemas are entirely undocumented, the tools do not declare what fields, structures, or data types they return. This is a critical gap that forces LLMs to guess what to extract from results. Error handling and recovery guidance are absent. No tool annotations (destructiveHint, readOnlyHint, idempotentHint) despite amber_run_protein_md being marked IRREVERSIBLE. The tools follow single-responsibility naming well but lack schema completeness and error-handling guidance expected of production tools.
Generate Amber input files without running simulations. Creates all input files for inspection and customization. Useful for dry-run preparation.
Prepare a protein system with tleap only. Includes solvation in water box and ion addition for neutralization. Outputs topology and coordinate files.
Run a complete MD simulation workflow for a single protein. Includes system preparation (solvation, ionization), energy minimization, heating (NVT), equilibration (NPT), and production MD (NPT).
Output schemas completely undocumented. Tools return structured data (MD trajectories, topology files, input files) but provide no schema documentation of response fields, structure, or data types. LLMs cannot determine what fields are available for downstream planning or composition.
No tool annotations despite irreversible operations. amber_run_protein_md is marked Risk: IRREVERSIBLE (runs production MD, consumes GPU time, generates irreversible simulation state) but lacks destructiveHint annotation. Agents cannot detect this is a high-consequence operation.
No error handling or recovery guidance. Tools have no documented error cases, retryability classification, or recovery steps. If a simulation fails (GPU OOM, invalid PDB, missing dependencies), LLM receives no guidance on what to do next.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 54 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 19 | - | v1 |
Parameter 'pdb_file' accepts file paths without validation constraints or guidance. Description states 'Path to the PDB file' but does not specify: absolute vs relative path, file existence, PDB format version, size limits, or valid protein structures. LLMs will guess.
Enum parameters not declared as enums. Parameters 'forcefield' and 'water_model' have constrained values (ff14SB|ff19SB, tip3p|opc|tip4pew) documented in descriptions, but are NOT declared as enum types in the schema. LLMs cannot auto-validate without text parsing.
Default parameters may cause unintended side effects. amber_run_protein_md defaults use_gpu to true, which will fail on CPU-only systems without explicit override. Default sim_time_ns=10.0 may not be suitable for all protein systems. Defaults should not silently degrade behavior.
job_name parameter is optional and defaults to PDB filename, but behavior on collision is undocumented. If two calls use the same PDB, will they overwrite results? Create separate job directories? LLM cannot reason about multi-call safety.