AI Assistants (MCP)

UploadKit ships an official Model Context Protocol server. Once you add it to your AI coding assistant, the assistant gains first-class knowledge of:

• Every one of UploadKit's 40+ React components — names, categories, inspirations, ready-to-paste usage
• How to scaffold the Next.js route handler at app/api/uploadkit/[...uploadkit]/route.ts
• How to wire <UploadKitProvider> into your root layout
• How to configure Bring Your Own Storage (S3, R2, GCS, Backblaze B2)
• The right install command for your package manager

No API key. No config. Runs locally via npx.

Install

claude mcp add uploadkit -- npx -y @uploadkitdev/mcp

{
  "mcpServers": {
    "uploadkit": {
      "command": "npx",
      "args": ["-y", "@uploadkitdev/mcp"]
    }
  }
}

{
  "mcpServers": {
    "uploadkit": {
      "command": "npx",
      "args": ["-y", "@uploadkitdev/mcp"]
    }
  }
}

{
  "mcpServers": {
    "uploadkit": {
      "command": "npx",
      "args": ["-y", "@uploadkitdev/mcp"]
    }
  }
}

{
  "context_servers": {
    "uploadkit": {
      "command": {
        "path": "npx",
        "args": ["-y", "@uploadkitdev/mcp"]
      }
    }
  }
}

{
  "experimental": {
    "modelContextProtocolServers": [
      {
        "transport": {
          "type": "stdio",
          "command": "npx",
          "args": ["-y", "@uploadkitdev/mcp"]
        }
      }
    ]
  }
}

Remote MCP

For clients that cannot run npx — ChatGPT custom connectors, Claude.ai's web UI, Smithery, and any hosted MCP host — UploadKit also exposes a remote Streamable HTTP endpoint:

https://api.uploadkit.dev/api/v1/mcp

Streamable HTTP, stateless, no auth required for v1. Tools are read-only and share the same 11-tool surface as the stdio server.

ChatGPT connector

1. Open ChatGPT → Settings → Connectors → Add custom connector.
2. Paste the URL above into the MCP Server URL field.
3. Save. The connector will handshake, discover the 11 tools, and become available in your chats.

Claude.ai web

Paste https://api.uploadkit.dev/api/v1/mcp as a custom MCP connector wherever Claude.ai's web UI exposes the "Custom connector" input. Until the UI is GA, the stdio install below remains the recommended path for Claude Desktop and Claude Code.

Smithery / mcp.so

Once listed on Smithery, users can install the remote endpoint from:

https://smithery.ai/server/io.github.drumst0ck/uploadkit

Test the remote endpoint

Health check (should return {"status":"ok","version":"0.5.2","tools":11}):

curl https://api.uploadkit.dev/api/v1/mcp/health

Full inspection with the official MCP Inspector (browse tools + call them interactively in a web UI):

npx @modelcontextprotocol/inspector --transport streamable-http https://api.uploadkit.dev/api/v1/mcp

Raw JSON-RPC handshake (no client needed):

curl -X POST https://api.uploadkit.dev/api/v1/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "MCP-Protocol-Version: 2024-11-05" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"curl","version":"0"}}}'

Self-host with Docker

The repository ships a minimal Dockerfile at the root that runs the stdio server inside a container. Most users never need this (the npx install is simpler), but it's useful for:

• Containerized CI pipelines that invoke MCP tools
• Corporate networks that can't reach registry.npmjs.org at runtime
• Air-gapped developer environments

# Clone + build
git clone https://github.com/drumst0ck/uploadkit.git
cd uploadkit
docker build -t uploadkit-mcp .

# Run — the container speaks JSON-RPC on stdio
docker run -i --rm uploadkit-mcp < request.json

To wire it into Claude Code or Cursor as a containerized server:

{
  "mcpServers": {
    "uploadkit": {
      "command": "docker",
      "args": ["run", "-i", "--rm", "uploadkit-mcp"]
    }
  }
}

Verify it's working

Ask your assistant:

"What UploadKit components are available?"

If the MCP server is connected, it will call the list_components tool and return the full catalog with categories (classic, dropzone, button, progress, motion, specialty, gallery).

What the server exposes

Tools

Resources

Example prompts

Once installed, you can speak to your assistant naturally:

Add an Apple-style animated dropzone to my profile page.

Set up UploadKit with Cloudflare R2 in BYOS mode.

What's the difference between UploadProgressRadial and UploadProgressOrbit?

Create a route handler that accepts up to 4 images of 8 MB each.

Show me every progress indicator UploadKit ships.

Why MCP

Most AI coding assistants can only see the files in your editor plus whatever they learned during pretraining. That means:

• They don't know about components shipped after their knowledge cutoff
• They invent props that don't exist (hallucinations)
• They write outdated setup code
• They can't scaffold correctly without guessing

The MCP server fixes all four. The assistant queries the authoritative source — UploadKit's own catalog and scaffolds — every time. New release? Update the server and the assistant knows the new components immediately.

Security & privacy

• The server runs locally on your machine via stdio.
• It does not make network requests, read your codebase, or collect telemetry.
• The only thing it ships is the component catalog (names, descriptions, usage examples) bundled at publish time.
• Source: github.com/drumst0ck/uploadkit/tree/master/packages/mcp

Troubleshooting

"Command not found: npx" — install Node.js 18+ from nodejs.org.

Server fails to start — run npx -y @uploadkitdev/mcp in a terminal manually. You should see the process hang (stdio waits for input). If it crashes instead, open an issue with the stderr output.

Tools don't show up in Claude Code — restart Claude Code after editing the config. Run claude mcp list to verify the server is registered.

"Outdated catalog" — the server is bundled with UploadKit at publish time. npx -y @uploadkitdev/mcp@latest pulls the newest version every time.

Resources

• Source: github.com/drumst0ck/uploadkit/tree/master/packages/mcp
• npm: @uploadkitdev/mcp
• Changelog: packages/mcp/CHANGELOG.md
• Glama listing: glama.ai/mcp/servers/drumst0ck/uploadkit (AAA)
• Official MCP Registry: io.github.drumst0ck/uploadkit on registry.modelcontextprotocol.io
• Issues: github.com/drumst0ck/uploadkit/issues