PaperDelta CLI
PaperDelta finds places in a paper that went stale after experiment results changed and gathers them into one report.
- 168 ★ on the repository
- Last commit
Minimal access
Why this level
- The tool reads the manuscript and data locally, and MCP mode writes nothing to paper files.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/paperdeltaDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The tool reads the manuscript and data locally, and MCP mode writes nothing to paper files.
This is third-party code. Review the repository files before installing.
What it does
PaperDelta, a local Python CLI, ties the numbers and claims in a manuscript to declared experiment data. If accuracy moves from 84.1 to 80.9 percent, the program shows where the abstract, table and appendix still repeat the old number and which conclusions no longer hold. Inputs: LaTeX, Markdown and Quarto, optionally Word and PDF. Arithmetic is exact, and checking needs no model key, GPU or TeX. For agents there are 22 optional MCP tools that do not modify paper files.
Who it is for. Researchers who rework experiments after writing a paper and fear leaving old numbers in the text.
Good fit when
- Rerunning an experiment changed a metric, and every place with the old number must be found.
- Before submission the paper is checked to confirm that the "outperforms baseline" claim still holds.
- An agent in Claude Code or Codex gathers edit candidates through MCP without touching files.
- The report opens offline as HTML and co-authors go through it place by place.
Not a fit when
- Experiment data is not recorded anywhere, while the tool needs a declaration of sources.
- Automatic text rewriting is wanted, but PaperDelta writes nothing to the paper files.
- The paper uses a format outside the list: static LaTeX, Markdown, Quarto, Word and PDF are supported.
Example request
Run a PaperDelta check on the paper and the data in `results/metrics.csv`, listing places with old numbers.Limitations
Python 3.11 or newer is required. Static markup is supported: dynamic LaTeX macros and complex includes are not parsed. Word and PDF reading come from the extra packages `paperdelta[docx]` and `paperdelta[pdf]`. MCP is enabled through `paperdelta[mcp]`. Automatic mapping remains experimental: in the real-model study of version 1.4 it reached no complete mappings out of nine. The tool shows places to review, and the decision about conclusions stays with the author. The latest version at review time is 2.0.0.
How to disable. Remove the package with `python -m pip uninstall paperdelta` and drop the `paperdelta mcp` launch from the agent config.
FAQ
Do I need a model key?
No. Checking runs locally with no model key, GPU or installed TeX.
Does the program change paper files?
No. The MCP tools read, check and prepare proposals, but they do not accept mappings or write paper files.
How do I enable PDF and Word support?
Install `paperdelta[pdf]` or `paperdelta[docx]`. The base install reads LaTeX, Markdown and Quarto.
Which paper formats are supported?
LaTeX, Markdown and Quarto work out of the box, while Word and PDF come through extra packages.
How many MCP tools does PaperDelta have?
Twenty-two optional ones. Four of them run batch candidate-selection sessions.
Discussion
No questions or reviews yet.
Related
CLI toolsMicrosoft's utility and MCP server that turn PDF, Word, Excel, PowerPoint, HTML and other files into Markdown for models
Anthropic skills
Skills
Anthropic's official skill examples: docx, pdf, pptx and xlsx documents, design, MCP builder and the Agent Skills spec
A skill that draws validated architecture, workflow, sequence and data-flow diagrams as one interactive HTML file
PaddleOCR agent skills
PaddleOCR Agent Skills
Two official PaddleOCR skills: text recognition from images and PDFs, and parsing complex documents into Markdown
MinerU is a CLI with a document library: search and revisit an already parsed report using its local cache.
Tesseract OCR is a CLI for multilingual scans: select language datasets and compare recognition on a sample page.
README in short
The README opens with an example where accuracy drops from 84.1 to 80.9 percent and shows a demo report. It continues with installation, a quick start, supported scope and the agent mode.