Find Technical Drawings by Part Number with AI: Search a PDF Drawing Archive

TL;DR: In most engineering companies the drawing archive is a folder tree of PDF drawings, exported from CAD, with the part number in the title block. AnythingMCP File Search indexes the text of every page, so Claude or ChatGPT can find a drawing by part number, drawing number, material or customer, open the title block, and answer with the file, the revision and its date. Native CAD files (DWG, DXF, STEP) are found by name and folder only: the index reads PDFs, not CAD geometry. Keep a PDF next to each drawing and both are covered.

The question that eats ten minutes

"Which drawing is part 4711-200, and what is the current revision?"

The answer is on page 1 of a PDF somewhere in Drawings/Conveyor/Assemblies/, unless it is in Drawings/Customer projects/2024/, or the file is named after the assembly and the part only appears in its parts list. Windows search does not look inside PDFs on a network share reliably, the PDM system is not open on every desk, and the colleague who knows the structure is on holiday.

What File Search reads in a drawing archive

A quick test for your own drawings: open a PDF and try to select the text in the title block. If you can, File Search can read it. Some CAD exports draw text as lines (often with CAD-specific stroke fonts); those PDFs have no text layer and behave like scans. Most CAD systems have an export option that keeps text as text, which also makes the PDFs smaller.

Set it up

1. Connect the archive as a File Search connector. Self-hosted, mount the drawing share read-only and choose Network folder (how). On the Cloud, expose it over SFTP, FTPS or WebDAV (NAS step by step).
2. Scan and select. The scan shows the number of files per folder and per type. Leave Old, Archive or Superseded folders unticked so withdrawn revisions are never quoted. If the archive is bigger than your plan, select the product lines people ask about most, or talk to us about Enterprise (up to 50,000 files on the Cloud, unlimited self-hosted).
3. Describe the conventions. Tell the AI what only your team knows, for example: "File names are <drawing number>_<revision>.pdf. The part number is in the title block field 'Part No.'. Revision letters increase A, B, C; the highest letter is current. Assemblies list their parts on the last page."
4. Name the tool prefix after the content, e.g. drawings, and give the connector to the roles of engineering, production and service.

anythingmcp.internal/connectors

How the assistant finds a drawing

“Which drawing is part 4711-200, and what is the current revision?”

File Search

Tools called

drawings_grepdrawings_finddrawings_read

1. _grep searches 4711-200 literally across all pages. It finds the title block of the part drawing and every parts list that uses the part, each with two lines of context, the page and the file date.
2. _find checks file names for the same number, which catches DWG and STEP files of the part and other revisions of the PDF.
3. _read opens page 1 of the best match to read the title block: title, material, revision, date.
4. The answer: "Part 4711-200 'Bearing plate' is drawing 4711-200_C.pdf in Drawings/Conveyor/Parts (revision C, 3 March 2026, material 1.4301). Revision B is in the same folder. The part is used in assemblies 4700-000 (pos. 3) and 4720-000 (pos. 7). A DWG with the same number sits next to the PDF."

Invented part numbers and drawings, for illustration.

Questions that work well

• "Find the drawing for part 4711-200 and give me the PDF."
• "Which assemblies use part 4711-200? List drawing number and position."
• "Show all drawings in Conveyor/Assemblies changed since March 2026."
• "Which drawings specify material 1.4404?"
• "Is there a test report for drawing 4720-000 revision C?"

The assistant hands over the drawing itself with _get_file: a temporary internal link that opens for signed-in members of your workspace.

Limits worth knowing

• No geometry. File Search does not interpret the drawing itself, only its text. "Find the bracket with two 12 mm holes" works only if that is written somewhere in text.
• Native CAD formats are not read. DWG, DXF, STEP and the like are listed by name and folder. Searching their content needs a PDF export (or a PDM connector).
• Revisions are decided by the AI from what it reads. Put the revision in the file name or the title block and say in the connector's instructions which one is current. Leave withdrawn revisions out of the index.
• Scans need OCR, which is part of Enterprise.

Plans and permissions

Drawing archives are large. On AnythingMCP Cloud, Starter covers 500 files and Team 2,000; Enterprise goes up to 50,000 files with a sync every 15 minutes, OCR for scanned drawings and folder permissions per role (for example: service sees Released/ only, engineering sees everything). Self-hosted Enterprise has no file limit and syncs every 5 minutes. Compare plans.

FAQ

Do our drawings leave the company? Self-hosted, the index stays on your server and only what a tool returns for a question (a match list, the text of a title block) reaches the AI provider. On the Cloud, the extracted text is stored in your workspace in Frankfurt; the files themselves are not kept.

Can it search DXF files? DXF is text. Not at the moment. DXF, DWG and other CAD formats are found by name and folder.

Does it replace our PDM or ERP? No. It makes the PDF archive searchable from the AI assistant your team already uses. Item master data, BOM structures and release status stay in the PDM or ERP, which you can connect as further connectors.

Related

• Search a Synology or QNAP NAS with Claude and ChatGPT, step by step
• Find an article number in PDF and Excel price lists
• Search an e-mail archive of .msg and .eml files with AI
• Folder permissions per role for AI file search
• Chat with your documents, RAG or File Search: which fits a company file server?
• Search your company file server with AI
• Network folder (self-hosted) in Claude, ChatGPT, Microsoft Copilot, Gemini