Repo to Prompt — Convert a Codebase Into One LLM Prompt

Modern context windows make "paste my whole repo into Claude" a real workflow. Claude Opus 4.7 and Gemini 2.5 Pro both handle 1 million tokens — enough for most medium-sized codebases. The challenge is fitting them under the cap and formatting the result so the LLM understands the file structure. freefilestoprompt.app handles both in three clicks.

The three-click repo-to-prompt workflow

Drag your repo folder into the drop zone. The browser reads it locally — no upload. All files become rows showing token counts.
Mark priorities. Pin must-keep architecture files (entry points, route definitions, schemas). Drop noise (lockfiles, build artifacts, fixtures, generated code). The rest gets high / medium / low.
Auto-fit and copy. The packer fills your budget greedily by priority. Output is one block with all files wrapped in delimiters and a directory tree at the top. Copy → paste into Claude / GPT / Gemini.

What gets packed

Any text-based file: .ts, .tsx, .js, .py, .go, .rs, .rb, .java, .kt, .swift, .c, .cpp, .h, .css, .html, .md, .json, .yaml, .toml, plus any extensionless text. Binaries (images, PDFs, archives, executables, fonts) are auto-skipped with a count of how many were filtered.

Choosing a target model

The dropdown includes 14 popular models with their context windows:

Claude Opus 4.7 / Sonnet 4.6 / Gemini 2.5 Pro / GPT-4.1 — all 1M tokens. Best for repo-scale prompts.
Llama 4 Scout — 10M tokens. Best for monorepos.
GPT-5 / GPT-5 mini — 400K tokens. Good for medium repos.
Grok 4 Fast — 2M tokens. Fast and cheap for repo Q&A.
Custom — enter any number.

Output format choice

XML format (<file path="...">CONTENT</file>) is the default and what Anthropic recommends for Claude. Markdown (### File: path + fenced code) is useful when you want the model to render the output back as Markdown. Plain (=== FILE: path ===) is the simplest separator. All three include an optional directory tree at the top so the model knows the file structure.

Try freefilestoprompt.app — Free, No Sign-Up

Drop files, set a target model, get one packed prompt. Runs entirely in your browser.

Open Files to Prompt →

Frequently Asked Questions

How many files can I pack at once?

Up to 200 files or 50 MB total combined, whichever comes first. Per-file cap is 10 MB.

What if my repo is larger than the context window?

Auto-fit drops lowest-priority files first. Mark the most important files as high priority, drop tests and fixtures, and the packer keeps what matters under the budget. Excluded files stay visible in the list, dimmed, so you can adjust priorities and re-fit.

Can I include uncommitted local changes?

Yes. freefilestoprompt.app reads files from your local filesystem, not from git. Whatever is on disk is what gets packed.

Does it preserve folder structure?

Yes. Dropped folders become files with their relative paths preserved, e.g. `src/api/auth.ts`. The optional directory tree at the top of the output reflects the structure.

What about binary files like images?

Auto-skipped. The drop handler detects binary extensions and files over 10 MB and excludes them with a toast notification of how many were filtered.

Can I include the diff or git history?

Not directly — freefilestoprompt.app reads files, not git metadata. Workaround: run `git diff > diff.txt` locally and drag that file in.