File Transfer (remote servers)

SkillFiles & storage

Move user files to or from a remote openstudio-mcp server — upload local models/measures/weather for server-side tools, or hand results back as download links. Use when a needed file is on the user's machine, or the user wants a file from the server.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the File Transfer (remote servers) skill

What this skill tells your AI

The instructions your AI receives, as published by natlabrockies/openstudio-mcp in .claude/skills/file-transfer/SKILL.md and read by ahel’s review.

Over HTTP the server runs on a different machine than the user. Every file tool (load_osm_model, apply_measure, change_building_location, import_floorspacejs) takes a server-side path, but the user's files are on their laptop. Never pass file bytes through tool arguments (base64 in-band is token-prohibitive and leaks bytes into context/logs) — move bytes out-of-band over signed HTTP routes on the same port.

Upload (get a local file onto the server)

  1. Check first: list_files() / list_uploads() — staged inputs (e.g. under /inputs) are already server-visible and need no upload.
  2. request_upload(filename="model.osm", size_bytes=<bytes>){file_id, upload_url, ...} (optional sha256 for integrity, kind="measure" for measure zips)
  3. Have the user PUT the raw bytes:
    curl --upload-file model.osm "<upload_url>"
    
  4. get_upload(file_id=<file_id>)server_path once ready (extracted_path for a .zip — ready for apply_measure(measure_dir=...))
  5. Pass that server path to load_osm_model / change_building_location / etc.

Download (hand a server file back to the user)

request_download(path=<server_path>) or request_download(run_id=<run_id>, bundle=True)download_url; the user fetches it:

curl -O -J "<download_url>"

For small text (logs, err files) prefer read_file / get_run_logs — no download needed.

Notes

  • Upload URLs are signed (HMAC, short TTL) and op/file-scoped; the bearer token never enters a URL. Uploads land in your private per-user area and count against a quota; archives are zip-bomb and path-escape guarded.
  • delete_upload(file_id=...) frees quota.
  • Shell-less clients: python -m mcp_server.tools.osmcp put <file> --kind measure and python -m mcp_server.tools.osmcp get <server_path> wrap the same two HTTP calls.

Signals

GitHub stars
33
Forks
8
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
file-transfer-natlabrockies
Source
github.com/natlabrockies/openstudio-mcp