Loading ToolBench
A materials science research assistant with access to multiple databases and tools for discovering materials based on requirements. Provides access to Materials Project (inorganic materials), PubChem (organic compounds), SureChEMBL (patent database), web search, and chemical structure visualization.
Materials Discovery Agent demonstrates solid tool definition quality with comprehensive parameter schemas and detailed descriptions. All 12 tools have explicit input schemas with type definitions and parameter descriptions. Tool names follow verb_noun conventions (search_*, get_*, lookup_*, visualize_*). Descriptions are substantive (100-400+ chars), exceeding the 10-1024 char baseline. However, output schemas are not documented, responses are inferred from context rather than formally specified. Error handling guidance is absent. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all tools being READ_ONLY. Parameter descriptions are thorough but some lack actionable constraints (e.g., 'similarity threshold' lacks guidance on typical values). No documented recovery paths for common failures.
Get frequency statistics for a chemical across the patent database.
Get statistical distribution of a Materials Project field to understand typical values.
Get full patent document content including title, abstract, and description. Use this to read patent text for detailed invention analysis. Patent ID format: {COUNTRY}-{NUMBER}-{KIND}, e.g., "WO-2020096695-A1"
Get all family members of a patent across different jurisdictions. Use this to assess global IP coverage for an invention. Shows same invention filed in multiple countries (up to 50 members).
Get comprehensive chemical data by SureChEMBL ID. Returns full molecular properties including: SMILES, InChI, molecular formula, molecular weight, LogP, polar surface area, heavy atoms, H-bond donors/acceptors, rotatable bonds, aromatic rings, drug-likeness: QED score, Lipinski violations, Rule of 3 compliance, patent frequency and organic/element flags.
Output schemas not documented. Tool descriptions state what is returned (e.g., 'Returns: SureChEMBL ID, SMILES, InChI...') but no formal JSON Schema for response structure. LLMs cannot plan downstream tool calls or extract fields reliably without documented output types.
No tool annotations despite all tools being READ_ONLY. Annotations (readOnlyHint, destructiveHint, idempotentHint) are missing from tool definitions. This prevents clients from optimizing caching, retry logic, and safety checks.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 64 | <=2025-11-25 | v2 |
Look up chemical by name in SureChEMBL to get SMILES, InChI, and basic properties. Returns: SureChEMBL ID, SMILES, InChI, molecular weight, patent frequency. Note: For comprehensive drug-likeness data, use lookup_chemical_by_id() with the returned ID.
Search Materials Project database for inorganic materials with computed properties. The Materials Project contains 170k+ experimentally known materials plus 380k+ from Google's GNoME dataset (predicted stable materials). Use tuple syntax [min, max] for numeric ranges.
Search SureChEMBL patent database for materials and applications.
Search PubChem for organic compounds, polymers, and molecular properties. PubChem contains 110M+ compounds with computed properties including molecular identifiers (SMILES, InChI, InChIKey), drug-likeness descriptors (XLogP, TPSA, H-bond counts), structural features (stereocenters, rotatable bonds), and 3D conformer data (volume, RMSD).
Find structurally similar chemicals in patents using SMILES notation.
Generate and display a chemical structure image from SMILES notation. The tool automatically displays the chemical structure image to the user. Simply mention that you've shown the structure: "I've displayed the structure above". You can analyze and describe what you see in the structure.
Search the web for general information, concepts, prices, or validation.
No error handling guidance. Tools lack descriptions of failure modes, recovery steps, or actionable error messages. E.g., 'search_patents' does not document what happens if query syntax is invalid or if no results are found.
Pagination not documented for list-returning tools. 'search_materials_project' and 'search_pubchem' accept 'limit' but no 'offset', 'page', or 'next_cursor' parameters. Large result sets may exceed context windows without pagination support.
Parameter constraints lack actionable guidance. 'similarity_threshold' (0.0 - 1.0) has no guidance on typical values or interpretation. 'num_results' (1 - 100) lacks explanation of performance trade-offs. Descriptions should include 'typical range' or 'recommended values'.