DFT AutoPilot — an MCP server that automates DFT computational chemistry (Quantum ESPRESSO / VASP / Gaussian) on your own HPC (SCNet, Slurm, or local).
Critical failures across all 19 tools. Descriptions are present but generic (avg ~80 chars, below the 194-char baseline). No parameter descriptions exist, tools accept arbitrary JSON with no guidance on what fields are required or what they control. No output schemas documented. No error handling guidance. Tool names lack clarity (e.g., 'run_pymatgen' is vague; 'extract_dft_results' vs 'extract_vasp_results' vs 'extract_gaussian_results' are near-duplicates). This server is not production-ready for agent use.
Cancel a submitted DFT computation job on HPC backend
Query the status of submitted DFT computation jobs
Generate Gaussian input files (.gjf, .com) from structure and calculation parameters
Generate Quantum ESPRESSO input files (scf.in, nscf.in, etc.) from structure and parameters
Generate VASP input files (INCAR, KPOINTS, POSCAR) from structure and parameters
Download computation results from HPC backend to local workspace
Parse and extract results from Quantum ESPRESSO output files (.out, .xml)
ALL 19 tools have empty input schemas: {type:object, properties:{}}. No parameters are defined, documented, or constrained. LLMs cannot determine what fields to pass or what they control.
No output schemas documented. Agents cannot plan downstream tool calls or extract required fields (e.g., job_id from submit_compute_job, which is needed by check_job_status and cancel_job).
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 33 | <=2025-11-25 | v2 |
Parse and extract results from Gaussian output files (.log, .fchk)
Parse and extract results from VASP output files (OUTCAR, vasprun.xml, DOSCAR)
Import crystal structures from various formats (CIF, POSCAR, XYZ, etc.)
Look up Hubbard U parameters for DFT+U calculations
Search and retrieve pseudopotential files for Quantum ESPRESSO calculations
Generate plots and charts from DFT calculation results (band structures, DOS, etc.)
Preview remote files on HPC backend with pagination support
Execute pymatgen operations for structure manipulation and analysis
Search academic literature for DFT-related papers and references
Search materials database for crystal structures and properties
Submit DFT computation jobs to HPC backend (SCNet, Slurm, or local)
Generate comprehensive reports from DFT calculation results
Three near-duplicate extraction tools (extract_dft_results, extract_vasp_results, extract_gaussian_results) with identical descriptions. LLMs cannot distinguish when to use each. Consider a single 'extract_results' tool with a 'software' enum parameter.
Generic tool name 'run_pymatgen' lacks clarity. Does it manipulate structures, compute properties, or something else? Rename to 'manipulate_structure' or 'analyze_structure' with a 'operation' enum parameter.
No error handling guidance. Tools like 'submit_compute_job' and 'cancel_job' are destructive/reversible but provide no recovery hints. Agents cannot determine if failures are retryable or require user intervention.