Loading ToolBench
Collection of MCP-compatible agent kits for document verification, identity checking, adverse media screening, AI text detection, and tender matching using the Stipple REST API
Stipple Kits exposes 16 tools with complete input schemas and descriptions. Naming follows verb_noun convention (verify_identity, screen_adverse_media, detect_ai_text). Descriptions are present and substantive (avg ~120 chars), explaining WHAT each tool does. However, parameter descriptions are sparse or missing context about constraints, formats, and dependencies. Output schemas are not documented, LLMs cannot predict response structure for chaining. Error handling is absent from tool definitions. No tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles (15 READ_ONLY, 1 WRITE). Security concern: create_key accepts email/name/label as parameters but no guidance on API key handling or secret injection. Composition is sound, tools are single-purpose and chainable (e.g., resolve_company → match_tenders). Baseline: avg tool description ~120 chars (within 10 - 1024 range), but param descriptions average ~40 chars and lack constraint details.
Check if a document has been inspected before (free cache check)
Check a pack of documents
Create a new API key
Detect probability that a document's prose is AI-written with linguistic tells
Extract specified fields from a document
Get a warrant by ID
Match tenders to a company
Output schemas not documented. LLMs cannot predict response structure (fields, types, pagination) for downstream chaining. E.g., verify_identity returns 'status'/'result', 'points_total'/'total_points', 'checks'/'documents_checked', inconsistent field names force LLM guessing.
Parameter descriptions lack constraint details. 'limit' has no min/max (tenders, match_tenders). 'scheme' enum is documented but no guidance on when to use afp_100_point vs austrac_safe_harbour. 'deep' boolean (verify_references) has no explanation of performance/cost tradeoff.
No tool annotations (readOnlyHint/destructiveHint/idempotentHint). 15 tools are READ_ONLY, 1 is WRITE (create_key), but MCP tool definitions lack these hints. Agents cannot infer safety/retry semantics without explicit annotations.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 69 | <=2025-11-25 | v2 |
Get pricing information for Stipple tools
Resolve a company query to structured company data
Corroborated adverse media and PEP/sanctions screening for a person or organisation based on name
Get list of tender sources
Search for tenders with optional query and jurisdiction filters
Get API usage information (requires API key)
Verify a document's authenticity with forensic signals and risk assessment
AFP 100-point identity check or AUSTRAC safe-harbour verification over a document set
Verify references from a document
Error handling absent from tool definitions. No guidance on retryability, user-fixable errors, or recovery steps. E.g., if verify_identity fails due to unsupported document type, LLM has no hint to call check_document or ask user for different docs.
create_key accepts email/name/label as parameters with no guidance on API key security, secret injection, or token handling. Tool description does not warn that API keys should never be logged or echoed to users.