Parse documents from Claude Code

Two quick commands. Ask Claude to parse any document — DOCX, PPTX, XLSX, PDF, and more.

Install the Skill

Run two commands inside Claude Code — add the AILANG Parse marketplace, then install the plugin:

# Run both inside Claude Code
/plugin marketplace add sunholo-data/docparse-skill
/plugin install ailang-parse@ailang-parse-marketplace

This registers the AILANG Parse skill in your Claude Code environment. Once installed, Claude automatically invokes it whenever you ask about document parsing, extraction, or conversion. No configuration files to edit.

The plugin connects Claude Code to the hosted AILANG Parse service at https://docparse.ailang.sunholo.com/mcp/connect/ and signs in with OAuth (see Authentication). The skill also knows the local docparse CLI, for documents that must not leave your machine. You just ask in plain English.

Usage Examples

After installing, just tell Claude what you need. The skill triggers on any document parsing task:

> Parse this contract and extract all track changes
> What tables are in quarterly_report.xlsx?
> Convert presentation.pptx to markdown
> Extract comments from legal_draft.docx with authors
> Parse invoice.pdf and return structured JSON
> Turn these release notes into a PowerPoint
> Write that summary up as a Word document

Claude handles file upload, format detection, output formatting, and error recovery. You get structured blocks back — ready to use in your code or pipe into another tool.

What the Skill Can Do

17 Input Formats

  • Office: DOCX, PPTX, XLSX, ODT, ODP, ODS
  • Text: CSV, Markdown, HTML, EPUB, EML, MBOX, TEX, RTF
  • PDF: text-layer extraction with pdftotext, no AI; AI only when you ask for it (scanned PDFs)
  • Images (AI): PNG, JPG

4 Output Formats

  • blocks — structured JSON with typed block ADT (Text, Heading, Table, Image, List, Section, Change, Audio, Video)
  • markdown — clean Markdown rendering
  • html — semantic HTML
  • a2ui — A2UI protocol for agent UIs

Key Features

  • Deterministic Office parsing — no AI needed, no network calls, no per-page billing
  • AI for images and scanned PDFs — pluggable models (Gemini, Claude, Ollama)
  • Track changes, comments, merged cells — structural features from direct XML access
  • Cost estimation before parsing — check AI vs deterministic and latency before committing
  • OAuth sign-in — sign in once in the browser; Claude Code holds the token, you never paste a key
  • Download links — generated Word, PowerPoint and Excel files come back as a one-hour download link

Authentication

The plugin signs in with OAuth — no API keys to copy or paste:

  1. Run /mcp, choose ailang-parse → Authenticate (or just ask for a parse; the first call that needs your account starts it)
  2. A sign-in page opens on docparse.ailang.sunholo.com, showing which app is asking
  3. Sign in with Google or GitHub and click Allow
  4. Claude Code keeps the token and renews it daily — every later request just works

The token is an ordinary AILANG Parse key on your account, listed on your dashboard as oauth: <app>, where you can revoke it. Usage counts against your plan like any other key; the Free tier gives you 1,000 requests/month to start.

Scripts, SDKs and headless agents use a dp_ API key instead: create one on the dashboard or with the device authorization flow (the skill's scripts/device-auth.sh, the SDKs, or the mcpAuth tool on the agent surface /mcp/), and set it as DOCPARSE_API_KEY. See two ways to sign in.

API Endpoints Available

The skill can access these endpoints on your behalf:

EndpointMethodPurpose
/api/v1/parsePOSTParse any document into structured blocks
/api/v1/estimatePOSTCheck cost and latency before parsing
/api/v1/formatsGETList all supported input/output formats
/api/v1/samplesGETTest with built-in sample files
/api/v1/capabilitiesGETFull capability manifest for agent discovery
/api/v1/healthGETService status and version

All responses include typed error codes with suggested_fix fields that Claude can act on directly. If a parse fails, Claude reads the fix and retries automatically.

MCP Server

AILANG Parse also works as an MCP (Model Context Protocol) server, exposing its parsing tools to any MCP-compatible AI agent:

# Coming soon
ailang-parse --mcp

The /api/v1/tools endpoint already serves tool definitions in Claude, OpenAI, and MCP formats. Any agent that speaks MCP can discover and call AILANG Parse without custom integration.

Try in Browser API & SDKs

Frequently Asked Questions

How do I install the AILANG Parse skill for Claude Code?

Inside Claude Code, run /plugin marketplace add sunholo-data/docparse-skill then /plugin install ailang-parse@ailang-parse-marketplace. Once installed, Claude Code can parse and generate documents during your coding session — extract tables, track changes and comments, or turn notes into a DOCX, PPTX or XLSX without leaving the terminal.

What can the AILANG Parse Claude Code skill do?

The skill parses and generates documents. Parsing: ask Claude to extract a specific table, list tracked changes with authors, or pull out comments. Generating: ask for a Word document, a deck or a spreadsheet — Claude writes Markdown and converts it to any of the 9 output formats. Same engine as the API.

Does the Claude Code skill send my documents to external servers?

The plugin's MCP tools use the hosted AILANG Parse service, so documents parsed through them are sent to it. You sign in once with OAuth (Google or GitHub) and Claude Code keeps the access token. For material that must stay on your machine, the skill uses the local docparse CLI instead: Office, text and PDF (pdftotext) files are parsed without uploading anything. See the plugin README for when to use which.

Can I use AILANG Parse with other AI coding assistants besides Claude Code?

Yes. AILANG Parse also runs as an MCP server that works with Cursor, Windsurf, and any MCP-compatible IDE tool. Additionally, the REST API can be called from any custom agent or script.

How do I report a bug or request a feature in AILANG Parse?

Just ask Claude to report it — the skill files it for you through the submit_feedback MCP tool (routed to the AILANG Parse maintainers), so you never have to leave your session to open a GitHub issue. It sends a title, description, and category (bug, feature, docs, or limitation); add a code snippet or contact address if you'd like a follow-up.