Loading ToolBench
MCP server for HumanPen - lower AI-detection scores, fix citations, condense and translate document files, keeping formatting intact.
HumanPen MCP demonstrates solid definition quality with complete schemas, detailed descriptions, and proper error handling. All 7 tools have explicit input schemas with typed parameters and descriptions. Tool names follow verb_noun convention (humanize_document, fix_citations, etc.). Descriptions are comprehensive (150-400 chars), explaining WHAT the tool does, WHEN to use it, and cost implications. However, output schemas are not formally documented in the source, and some parameter relationships (e.g., mutually exclusive min_words/max_words/report_path) could be clearer. Error handling includes actionable guidance (e.g., 'Call check_job with this job_id'). Security is strong: API key injected via environment, no credentials in parameters. Tool composition is clean: each tool has one responsibility, and output includes IDs for chaining (job_id for check_job). Naming is clear and consistent.
Check the status of a running or finished job by its id. Does not cost credits.
Shorten a .docx or .pptx by a target percentage or word count, keeping the most important points. Saves the result next to the source and returns its path. COSTS CREDITS at 100 per 1,000 words processed (10 minimum) - say so and get agreement first.
Reformat citations in a .docx or .pptx to a target style (APA, MLA, IEEE, Harvard, GB/T 7714, or Chicago). Saves the result next to the source and returns its path. COSTS CREDITS at 100 per 1,000 words processed (10 minimum) - say so and get agreement first.
Check your HumanPen account's credit balance. Does not cost credits.
Rewrite a .docx or .pptx so it reads as human-written and scores lower on AI detectors, keeping meaning, citations, tables and layout. Optionally pass a Turnitin/iThenticate AI report to rewrite only the passages it flagged. If a fresh report still flags the result, run this tool again on the rewritten file with that new report - only the still-flagged passages are touched, and balanced remains the right strategy for the second pass. Saves the result next to the source and returns its path. COSTS CREDITS at 100 per 1,000 words processed (10 minimum) - say so and get agreement first.
Output schemas not formally documented. Tool descriptions explain what is returned (job status, output_path, etc.) but no JSON Schema for response structure is visible in source. LLMs cannot reliably parse or chain results without explicit output type definitions.
Mutually exclusive parameters not clearly documented. humanize_document has min_words/max_words that cannot combine with report_path or segments, but this constraint is buried in parameter descriptions rather than stated upfront. LLMs may pass invalid combinations.
target_lang parameter in translate_document lacks enum constraint. Description provides examples ('es', 'fr', 'zh') but no formal enum list. LLMs may hallucinate invalid language codes like 'english' or 'spanish' instead of ISO 639-1 codes.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 68 | 2026-07-28+ | v2 |
Parse a Turnitin or iThenticate AI-detection report PDF and return what it found: the AI percentage, the flagged passages, and the problem types. Does not cost credits.
Translate a .docx or .pptx to another language, keeping formatting, tables, images and layout. Saves the result next to the source and returns its path. COSTS CREDITS at 100 per 1,000 words processed (10 minimum) - say so and get agreement first.
Async job handling requires explicit polling. Tools return job_id when wait_seconds budget expires, but no automatic retry or background polling is offered. Agents must remember to call check_job, which is error-prone. Consider offering a 'wait_for_completion' parameter or automatic polling wrapper.
Cost warnings are verbal, not enforced. Descriptions state 'COSTS CREDITS at 100 per 1,000 words, say so and get agreement first', but no dry-run or cost_estimate parameter exists. Agents cannot check cost before committing, risking unexpected credit depletion.