Exposes PaperBanana's core functionality as MCP tools usable from Claude Code, Cursor, or any MCP client. Generates publication-quality academic illustrations (methodology diagrams, statistical plots) with AI-powered refinement.
PaperBanana-CN provides 11 tools with complete input schemas and descriptions, but exhibits significant quality gaps. All tools have descriptions (10-200 chars, baseline 194), and parameters include types and descriptions. However, output schemas are entirely undocumented, no tool specifies what fields or structure the response contains, forcing LLMs to guess downstream data shapes. Tool naming is inconsistent: 'generate_diagram', 'continue_run', 'continue_diagram', 'continue_plot' create ambiguity (why three 'continue' variants?). Parameter descriptions lack actionable constraints (e.g., 'iterations' has no min/max bounds; 'aspect_ratio' lists supported values but no guidance on default behavior). Error handling is absent, no recovery guidance, no categorization of retryable vs. fatal errors. Security: 'config' and connection profile parameters suggest file paths and credentials may be exposed as tool inputs rather than injected server-side. No tool declares permissions or audit requirements. Composition: batch_diagrams and batch_plots duplicate single-item tools without clear differentiation. Overall, the server is functional but lacks production-grade polish.
Batch methodology diagrams from a YAML/JSON manifest.
Batch statistical plots from a YAML/JSON manifest.
Continue a prior methodology run (more refinement / feedback).
Continue a prior statistical-plot run.
Continue refinement for a prior diagram or plot run.
Download the PaperBananaBench reference set (~298 examples).
Evaluate a generated diagram against a reference.
No output schemas documented. Tools define inputs but never specify response structure (fields, types, pagination). LLMs cannot plan downstream calls or extract required data.
Ambiguous 'continue' tool naming. Three separate tools (continue_run, continue_diagram, continue_plot) with overlapping semantics. LLMs cannot distinguish when to use each. Consolidate or clarify in descriptions.
Numeric parameters lack bounds. 'iterations' (default 3, max 30 in code) has no min/max in schema. 'aspect_ratio' lists enums but no guidance on invalid values. LLMs may pass absurd values.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | <=2025-11-25 | v2 |
Evaluate a generated plot against a reference.
Generate a publication-quality methodology diagram from text.
Generate a statistical plot from JSON data.
Full-paper figure package (plan + optional generation).
Credentials and file paths exposed as parameters. 'config', 'vlm_connection', 'image_connection' suggest file paths or credential references passed as tool inputs. Should use server-side secret injection.
No error handling guidance. Tools do not document what errors are retryable, what user action is needed, or how to recover. LLMs receive raw failures with no next-step guidance.