Using Claude Cowork to Create RFPs

Anthropic launched Claude Cowork in January 2026, a feature in the Claude Desktop app that gives Claude direct access to files and folders on your local computer. Over the weekend, we tested Cowork across several procurement use cases. Today we're writing about one of them: creating an RFP.

We gathered scope documents, existing contracts, and a standard RFP template, dropped them into a folder, pointed Cowork at that folder, and typed a prompt. Twenty minutes later, a complete RFP sat in the output folder, ready for review.

That same document would have taken several hours to assemble manually.

The workflow is simple: give Cowork access to a folder, describe what you need, and it reads your documents, asks follow-up questions, and produces the final RFP document, all saved locally. No uploading into a chat window. No copy-pasting. No hitting attachment size limits.

For category managers still building RFPs manually, this is the tool worth paying attention to.

Who Benefits Most

If your organization uses a dedicated sourcing platforms like Jaggaer, Coupa, SAP Ariba, or similar, this article isn't suggesting you replace it. Those tools manage the full procurement lifecycle and they're effective at it.

This is for two groups:

Teams building RFPs manually: You assemble documents in Word, track responses in spreadsheets, and email suppliers directly. Cowork handles the document creation step that eats most of your time.

Teams on legacy SaaS platforms: If your sourcing tool hasn't added AI features, Cowork can sit alongside it, generating the RFP document while your existing platform manages the bidding and evaluation process.

The Actual Workflow

Here's what the process looked like when we tested it:

Step 1: Prepare your folder. We created a dedicated folder and added everything relevant — scope documents, an existing contract for similar services, technical specifications, and our team's standard RFP template. Budget 30 to 60 minutes for this step, depending on how scattered your files are.

Step 2: Connect Cowork. In the Claude Desktop app, we pointed Cowork to the folder so it could read the contents and save new files there.

Step 3: Prompt. We described the RFP we needed — the category, supplier requirements, timeline, and evaluation approach. One prompt, plain language.

Step 4: Follow-up questions.** Claude asked several clarifying questions — details about evaluation weighting, submission format preferences, specific compliance requirements. We answered them directly in the chat.

Step 5: Walk away. Claude got to work. We didn't sit and watch. Twenty minutes later, the finished RFP appeared in the output folder.

Total time from document gathering to finished draft: roughly an hour. Compare that to the typical manual process — easily three to five hours for a moderately complex RFP — and the math speaks for itself.

What Makes Cowork Different

Most AI tools require you to copy-paste content into a chat window or upload files one at a time. Cowork skips that friction entirely. It connects directly to folders on your local machine.

You point it at a folder. It reads everything inside, contracts, scope documents, specifications, past RFPs. Then it produces new files, saved right back to a folder you specify.

No uploading. No copy-pasting. No hitting file size limits halfway through.

Anthropic's latest models also handle significantly longer tasks than earlier versions. Cowork can process larger document sets and produce more comprehensive output without running into the context limitations that made earlier AI tools impractical for complex RFP work.

Getting the Best Results

The quality of Cowork's output scales directly with the context you provide. A bare-bones prompt produces a generic RFP. A prompt backed by existing contracts, detailed scope documents, and specific evaluation criteria produces something much closer to final.

Two tips that made a real difference in our test:

Provide your existing RFP template: If your team uses a standard format, include it in the folder. Cowork will structure its output to match your headers, section order, and formatting saving you the reformatting step entirely.

Include historical contracts: Past agreements give Claude reference points for pricing language, SLAs, and contractual terms your organization has already negotiated.

Make It Repeatable with Custom Skills

Here's where the efficiency compounds. Cowork supports custom skills — reusable prompts you create once and trigger with a single command.

Think of it as building an "RFP Generator" button. You write a detailed prompt that captures your organization's RFP requirements, preferred structure, evaluation criteria, and formatting standards. Save it as a skill. The next time someone on your team needs an RFP, they point Cowork at a folder of source documents, type the skill command, and let it run.

No rewriting the prompt. No remembering which instructions produced the best results last time. One command, consistent output.

Implementing This for Your Team

You don't need to overhaul your procurement process to start. Pick one upcoming RFP — ideally a category your team has handled before, where you already have contracts and templates on hand.

Have one category manager run the full workflow: gather documents, point Cowork at the folder, prompt, review the output. That first test takes about an hour and gives your team a concrete sense of what Cowork produces and where human review still matters.

Once you're satisfied with the output quality, build a custom skill that captures your organization's RFP standards. That skill becomes the repeatable process for the rest of the team — no training required beyond "drop your docs in a folder and run this command."

From there, expand to other procurement document types. We tested Cowork across several use cases over the weekend, and RFP creation was just one of them. The same folder-based workflow applies to scope of work documents, supplier questionnaires, and evaluation templates.

Let's Talk

Want the detailed step-by-step guide, including the exact prompts we used? Reach out to sk@moleculeone.ai or dm@moleculeone.ai. We'd also love to hear from teams already doing this or chat if you're just getting started.

Written by humans refined by agents