transfa MCPサーバー

---

$ tf upload model.gguf
✓ Uploaded  model.gguf  (4.2 GB)  →  https://transfa.sh/f/xK9mRp
  SHA-256    a3f8c2...d91b
  Expires    in 7 days

---

Why transfa?

Most file-sharing tools are built for humans clicking through UIs. transfa is built for the terminal — designed to be called from shell scripts, CI pipelines, and AI agents (Claude, GPT, Cursor, etc.) that need to move files without friction.

• No signup required — just install and upload
• Any format, any size — up to 100 GB; ML models, archives, binaries, code, media
• Built for agents — JSON API, SHA-256 checksums, idempotent uploads
• Password protection, download limits, TTL — full control over every link
• 100+ formats detected — MIME type auto-detection including .gguf, .safetensors, .parquet, .ipynb, and more

---

Install

Node.js / CLI

npm install -g transfa

Python SDK

pip install transfa

MCP server (Claude, Cursor, any MCP-compatible agent)

npx -y transfa-mcp

Or use the raw install script:

curl -fsSL https://transfa.sh/install | sh

---

Quick start

# Upload a file (no account needed)
tf upload photo.jpg

# Upload with custom TTL and password
tf upload secret.zip --ttl 24h --password hunter2

# Upload and pipe the URL to clipboard
tf upload bundle.tar.gz | grep url | awk '{print $2}' | pbcopy

# Download a file
tf download https://transfa.sh/f/xK9mRp

# List your uploads
tf list

# Delete an upload
tf delete xK9mRp

---

GitHub Actions

Upload build artifacts, coverage reports, and any CI output straight from your workflow:

- uses: colapsis/transfa-action@v1
  id: upload
  with:
    file: ./dist/report.pdf
    api-key: ${{ secrets.TRANSFA_API_KEY }}

- run: echo "${{ steps.upload.outputs.agent-link }}" >> $GITHUB_STEP_SUMMARY

All five outputs are available after the step: id, agent-link, human-link, sha256, expires-at.

See colapsis/transfa-action for the full input reference and more examples (password-protected links, self-hosted instances, single-download limits).

---

API

transfa is fully REST. Every operation the CLI does, you can do with curl or any HTTP client.

Upload

curl -X POST https://transfa.sh/api/upload \
  -H "Authorization: Bearer $TF_KEY" \
  -F "file=@model.gguf" \
  -F "ttl=7d"

{
  "id": "xK9mRp",
  "url": "https://transfa.sh/f/xK9mRp",
  "download_url": "https://transfa.sh/api/download/xK9mRp",
  "filename": "model.gguf",
  "bytes": 4512345678,
  "sha256": "a3f8c2...d91b",
  "expires_at": "2026-05-21T12:00:00.000Z"
}

Upload options (form fields or headers):

Field	Header	Description
ttl	X-Transfa-TTL	Expiry: 1h, 24h, 7d, 30d
password	—	Password-protect the download link
max_downloads	—	Burn after N downloads
filename	X-Transfa-Filename	Override the stored filename

Download

# Direct download (no auth required)
curl -L https://transfa.sh/api/download/xK9mRp -o model.gguf

# Password-protected
curl -L "https://transfa.sh/api/download/xK9mRp?password=hunter2" -o secret.zip

File info

curl https://transfa.sh/api/download/info/xK9mRp

{
  "id": "xK9mRp",
  "filename": "model.gguf",
  "bytes": 4512345678,
  "sha256": "a3f8c2...d91b",
  "mime_type": "application/octet-stream",
  "download_count": 3,
  "has_password": false,
  "expires_at": "2026-05-21T12:00:00.000Z",
  "active": true
}

List uploads

curl https://transfa.sh/api/upload \
  -H "Authorization: Bearer $TF_KEY"

Delete

curl -X DELETE https://transfa.sh/api/upload/xK9mRp \
  -H "Authorization: Bearer $TF_KEY"

---

Supported formats

Over 100 file types with correct MIME detection — including types not in standard MIME databases:

Category	Formats
ML models	.gguf .ggml .safetensors .onnx .pt .pth .pkl .ckpt .tflite .mlmodel .lora
Data science	.parquet .arrow .feather .h5 .hdf5 .npz .npy .lance .duckdb .ipynb
Code	.py .rs .go .ts .kt .swift .scala .cu .sol .vy .elm .zig
Archives	.zip .tar .gz .bz2 .xz .7z .zst
3D / Design	.glb .gltf .obj .stl .usdz .blend .fig .sketch .psd
Media	.avif .webp .heic .jxl .opus .flac .webm .av1
Config	.toml .hcl .tf .tfvars .nix .dhall .lock .env
Everything else	.wasm .sqlite .db .pem .crt .p12 + all standard types

Any other format is accepted as application/octet-stream — nothing is blocked.

---

Plans

	Guest	Free	Pro	Team
Max file size	10 MB	500 MB	50 GB	100 GB
Uploads / day	5	20	500	5,000
Max TTL	24h	48h	30 days	180 days
Storage	—	—	Unlimited	Unlimited
Price	Free	Free	$12/mo	$48/mo
Trial	—	—	3-day free trial	3-day free trial

→ See full pricing

---

MCP server (Claude, Cursor, and any MCP-compatible agent)

transfa ships an MCP server that lets Claude, Cursor, and any MCP-compatible agent upload and share files autonomously — no shell commands, no infrastructure setup.

npx -y transfa-mcp

Claude Desktop config

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "transfa": {
      "command": "npx",
      "args": ["-y", "transfa-mcp"],
      "env": {
        "TRANSFA_API_KEY": "your-api-key"
      }
    }
  }
}

The API key is optional — the server works in guest mode without one (10 MB / 24h limit).

Available MCP tools

Tool	Description
upload	Upload a file from the local filesystem. Returns agent_link (direct URL), human_link (share page), and sha256. Accepts run_id, step, consumer, intent for provenance.
file_info	Get metadata about an upload — filename, size, SHA-256, expiry, download count, provenance fields.
list_uploads	List recent uploads (requires API key).
delete_upload	Delete an upload immediately.
run_artifacts	Get all files uploaded under a run_id — the full provenance manifest for a pipeline run or agent session.

Example agent workflow

When Claude has transfa as an MCP tool, it can:

1. Generate a report → call upload → get a link → paste the link in the conversation
2. Pass a file to another agent by sharing the agent_link
3. Clean up with delete_upload when done

---

Python SDK

import transfa

# Upload
result = transfa.upload("model.pt", ttl="24h", run_id="run-42", artifact=True)
print(result.url, result.sha256)

# Download (SHA-256 verified)
transfa.download(result.id, output="model.pt")

# Provenance manifest for a run
manifest = transfa.run_artifacts("run-42")

# Async
async with transfa.AsyncClient() as client:
    result = await client.upload("model.pt")

See python/README.md for the full API reference.

---

Use with AI agents (script/subprocess)

transfa is also designed to be called from shell scripts, CI pipelines, and agents that prefer subprocess calls:

import subprocess, json

result = subprocess.run(
    ["tf", "upload", "output.csv"],
    capture_output=True, text=True
)
data = json.loads(result.stdout)
print(data["url"])  # https://transfa.sh/f/xK9mRp

Or use the REST API directly — no SDKs, no auth flows, just HTTP.

---

Self-hosting

git clone https://github.com/colapsis/transfa.git
cd transfa
cp .env.example .env          # fill in your keys
npm install --prefix server
npm install --prefix cli
npm run build --prefix frontend
pm2 start ecosystem.config.cjs

Requirements: Node.js 18+, nginx (for SSL/proxy)

See nginx/transfa.conf for a production-ready nginx config.

Environment variables

Variable	Description
PORT	Server port (default: 3001)
BASE_URL	Public URL e.g. https://transfa.sh
STRIPE_SECRET_KEY	Stripe secret key for billing
STRIPE_WEBHOOK_SECRET	Stripe webhook signing secret
STRIPE_PRO_PRICE_ID	Stripe price ID for Pro plan
STRIPE_TEAM_PRICE_ID	Stripe price ID for Team plan

---

Security

Found a vulnerability? Please email tansfa.sh@gmail.com or see SECURITY.md.

Do not open a public issue for security reports.

---

License

MIT — © 2026 transfa contributors