Claude Code OCR: how to pull the tables out of a PDF

Can Claude Code read the PDF by itself?

Yes, up to a point. Claude Code can open a PDF or an image and tell you what it sees, and for a glance at a short document that is enough. What it cannot do is hand you the numbers as numbers. Ask it for the price table and you get a paragraph that describes the table; ask it for a 40-page filing and it runs out of room before the end. OCR — optical character recognition, software that turns a picture of text into actual text — is a different job, and Claude Code has no OCR service of its own.

What does OCR give me that reading does not?

A file. The document OCR service returns the page as text with the tables kept as tables — every row, every cell, in order — plus any images it lifted out of the document. Because the result is a file in your project, the follow-up is ordinary local work the agent does for free: turn a table into CSV, search the text, compare this quarter's sheet with last quarter's. That is what makes a 30-page report one call instead of 30 screenshots.

What does it cost, and how do I run it?

From $0.01 per item, settled by actual usage with a cap per call, so a one-page sheet and a forty-page filing do not cost the same and the receipt shows the settled amount. You run it by asking; the session below is the whole thing.

SKU,Size (EU),Size (US),Colour,Units per carton,Wholesale / pair,Suggested retail
EWS-GUM-38,38,6,White / gum sole,12,$31.50,$89.00
EWS-GUM-39,39,6.5,White / gum sole,12,$31.50,$89.00
EWS-GUM-40,40,7,White / gum sole,12,$31.50,$89.00
EWS-GUM-41,41,8,White / gum sole,12,$32.00,$89.00
EWS-GUM-42,42,9,White / gum sole,12,$32.00,$89.00
EWS-GUM-43,43,10,White / gum sole,12,$32.50,$89.00
EWS-GUM-44,44,11,White / gum sole,12,$32.50,$89.00
EWS-GUM-45,45,12,White / gum sole,6,$33.00,$89.00

That is a real session, unedited. The PDF was a one-page sample supplier sheet for the sneaker in our demo series — three tables: sizes and prices, lead times, materials. All three came back with every cell in place; the CSV above is the first one, converted locally. The run went through Actionway's document OCR capability; the Skill installed with the CLI is what told Claude Code the service existed, what it costs, and how to hand it a file.

What exactly comes back?

Text in reading order, page by page, with each table returned as table markup — rows and cells, not a picture of a grid — and any embedded images saved alongside. Headings survive as headings. The agent writes all of it into outputs/ and then does whatever you asked with it: CSV for one table, a Markdown file for the whole document, a comparison against a file you already have. Nothing in that second step is charged.

What can I feed it?

A PDF of up to 40 pages, an image — screenshot, photo of a page, scan — or a generic file, each by its web address; a local file is registered as an asset first, which the agent does for you. A financial statement comes back with the columns intact, ready to cross-check against the stock and company data services; a scanned contract comes back searchable. There is no subscription; one balance covers every capability in the catalog.

How does the session decide what to run?

Before replying, the agent checked the capability index the Skill keeps on disk, so the first message already carried the plan: the service, its price, and the honest alternative of just looking at the file. The list is neutral — no default, no recommendation. Picking an option is the approval: the call runs at the shown price, and if the wallet cannot cover it the agent stops before anything is charged and gives you a top-up link with the amount prefilled. OCR runs as a job, so the agent waits on it rather than you polling.

Set it up once

npm install --global @actionway/cli@latest
actionway init
mkdir -p "$HOME/.claude/skills"
cp -R "$HOME/.actionway/skill/actionway" "$HOME/.claude/skills/actionway"

You need Node.js 20 or newer and an Actionway account; after sign-in the agent tells you your wallet balance and, if it is empty, links you straight to top-up. Prefer a plugin or a Remote MCP server to the command line? Both end at the same catalog and the same wallet — the get started page lists the steps per client side by side.