AI coding agents (Claude Code, Codex, Cursor, and others) read code brilliantly — but binary office files like .docx and .pptx have always been painful. anydoc solves it as an Agent Skill: one command and your agent can "read" your documents.

Install

npx skills add firecrawl/anydoc

Once installed, skills-protocol agents will have a document-conversion capability in your project, invoked automatically whenever they hit an office file.

How to use it

No extra configuration. When your agent encounters a document, it follows this chain:

1. Notices the target is a binary file like .docx / .pptx / .pdf
2. Calls the anydoc skill to convert it to Markdown
3. Answers, edits, or summarizes based on the converted content

You can also ask explicitly: "Use anydoc to convert spec.docx to Markdown and summarize the key points."

Typical scenarios

Feed requirement docs straight to your agent

Drop PRDs (PRD.docx) or API docs into the project and let the agent read them — no manual Markdown conversion or copy-paste.

Batch contract / terms analysis

Hand an agent a folder of PDF contracts: it converts each one, extracts key clauses, and diffs them — all locally, documents never leak.

Proposals / report summaries

.pptx decks and .xlsx datasets get converted to Markdown, then the agent writes summaries, drafts presentation copy, or pulls data insights.

Under the hood

An Agent Skill is essentially a pre-configured tool for the agent: the anydoc CLI or binding is wrapped as an invocable skill command. Conversion still runs locally (no external service), so security properties match using anydoc directly.

Compatibility notes

• Requires an AI coding tool with skills-protocol support: Claude Code, Codex, Cursor, and other mainstream tools
• Skill marketplace page: skills.sh/firecrawl/anydoc
• Scanned PDFs remain unsupported (they need OCR) — consistent across every binding

One-liner

npx skills add firecrawl/anydoc — after that, your AI agent can read any document.

Contributors: souvc