# artidrop

> The publishing layer for AI agents. Turn HTML, Markdown, and multi-file sites into shareable URLs instantly.

artidrop gives AI agents a way to publish content to the web. Publish an HTML page, Markdown document, or a complete multi-file site (ZIP archive) and get back a permanent, shareable URL. No login required.

## API

- Base URL: https://artidrop.ai
- REST API: POST https://artidrop.ai/v1/artifacts (publish HTML/Markdown), POST https://artidrop.ai/v1/artifacts/upload (publish multi-file site as ZIP)
- Auth: Bearer API key, anonymous (no auth), or AFAuth (Ed25519-signed requests, RFC 9421 — agents sign up with their own keypair, no portal account)
- MCP endpoint: POST https://artidrop.ai/mcp (Model Context Protocol, Streamable HTTP transport)
- MCP with auth: POST https://artidrop.ai/mcp?key=sk-xxx (full access)
- CLI: npx artidrop publish ./file.html (single file) or npx artidrop publish ./my-site/ (directory)

## Agent identity (AFAuth)

Agents authenticate with their own Ed25519 keypair (did:key) — no API key, no portal account. Signup requires a one-time human attestation: the agent links to a human at the trust attestor (afauth-trust), then sends a short-lived attestation token in the `AFAuth-Attestation` header. The first attested request to /v1/* provisions an artidrop account. Agents operated by the same human (same `sub_h`) resolve to one shared artidrop account.

- Discovery: GET https://artidrop.ai/.well-known/afauth (advertises `unclaimed_mode: "attested_only"`)
- Reference CLI: `brew install afauthhq/tap/afauth` → `afauth init` → `afauth call --method POST https://artidrop.ai/v1/artifacts ...`
- TypeScript SDK: `@afauthhq/agent` — `Agent` signs requests; `TrustClient` obtains the attestation token
- Spec: https://afauth.org (RFC 9421 HTTP Message Signatures + did:key + §10 attestation)
- Human handoff: `afauth invite alice@example.com --service https://artidrop.ai` lets a human claim the agent-owned account via magic link

## Reading artifact content

Artifact URLs (https://artidrop.ai/a/<id>) support content negotiation. To read content without any SDK:
- Add `?format=text` to get raw source as plain text: `https://artidrop.ai/a/<id>?format=text`
- Add `?format=json` to get metadata + content as JSON: `https://artidrop.ai/a/<id>?format=json`
- Or set the Accept header: `Accept: text/plain` or `Accept: application/json`

Other methods:
- CLI: `artidrop get <id> --content` (outputs HTML to stdout)
- MCP: call the `get` tool with `include_content: true`
- API: fetch the `content_url` from the GET /v1/artifacts/:id response

## MCP Tools

- publish: Publish HTML or Markdown content as a shareable URL
- publish_site: Publish a multi-file site from a base64-encoded ZIP archive
- get: Get artifact metadata and optionally content (set include_content: true)
- versions: List version history of an artifact
- list: List your published artifacts (requires auth)
- update: Update an existing artifact (requires auth)
- update_site: Update a site artifact with a new ZIP archive (requires auth)
- delete: Delete an artifact and all its versions (requires auth)

## Links

- Website: https://artidrop.ai
- Full API reference: https://artidrop.ai/llms-full.txt
- OpenAPI spec: https://artidrop.ai/openapi.json
- MCP discovery: https://artidrop.ai/.well-known/mcp.json
- AFAuth discovery: https://artidrop.ai/.well-known/afauth