Connect Claude

Both MCP servers are live on Cloudflare Workers, so connecting is just pasting a URL. The two server URLs, used on every tab below:

• Procurement MCP (operate the lifecycle, token-gated, private slot):

  https://partifact-procurement-mcp.thanhvuttv.workers.dev/mcp/partifact-public-demo/<your-slot>

• Docs MCP (read the docs as tools, public, no token):

  https://partifact-docs-mcp.thanhvuttv.workers.dev/mcp

Pick your client.

claude.ai

Open Settings → Connectors → Add custom connector, paste a server URL above, and save. Start a fresh chat with the connector enabled.

Sanity check: on the Docs MCP, ask the agent to run search_docs("verify webhook signature") (it should return the Webhooks page); on the Procurement MCP, ask it to call get_started.

Note

Custom connectors are on claude.ai’s paid plans (Pro / Team / Enterprise) and may be labelled slightly differently per plan. If you don’t see them, use the Claude Desktop or Claude Code tab.

Claude Desktop

On a paid plan, the simplest route is the connectors UI: Settings → Connectors → Add custom connector, paste a server URL above, save, and fully restart Claude Desktop.

No connectors UI on your plan? Bridge the remote servers with mcp-remote in claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/claude_desktop_config.json; Windows: %APPDATA%\Claude\claude_desktop_config.json):

{
  "mcpServers": {
    "partifact-procurement": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://partifact-procurement-mcp.thanhvuttv.workers.dev/mcp/partifact-public-demo/<your-slot>"
      ]
    },
    "partifact-docs": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://partifact-docs-mcp.thanhvuttv.workers.dev/mcp"]
    }
  }
}

Keep only the server(s) you want, save, and fully restart Claude Desktop. Sanity check: ask the agent to search_docs("verify webhook signature") on the Docs MCP, or get_started on the Procurement MCP.

Claude Code

Add each server over HTTP:

# Procurement MCP, operate the lifecycle (token + slot in the URL)
claude mcp add --transport http partifact-procurement \
  "https://partifact-procurement-mcp.thanhvuttv.workers.dev/mcp/partifact-public-demo/<your-slot>"

# Docs MCP, read the docs as tools (public, no token)
claude mcp add --transport http partifact-docs \
  "https://partifact-docs-mcp.thanhvuttv.workers.dev/mcp"

Prefer the token out of the URL? Pass it as a header and keep the slot in the path: --header "Authorization: Bearer partifact-public-demo". To share with a repo, write a project-scoped .mcp.json (--scope project).

Sanity check: claude mcp list shows each server plus a connection check; inside a session, /mcp lists the connected tools.

The slot model

<your-slot> is any name you choose ([a-z0-9_-]). It selects your own private sandbox, seeded with the same 2019 Toyota Corolla repair job (CCC-2026-04817). Two people on two slots never touch each other’s data; pick a new name any time for a clean slate, and idle slots self-reset after about an hour. The public token partifact-public-demo only gates the hosted endpoint against random traffic. Full detail on Credentials and tokens.

Gotchas

• A wrong or missing token on the Procurement MCP returns 401 unauthorized.
• Omitting the slot (.../mcp/<token> with no trailing name) drops you in a single shared sandbox: fine for a quick look, but use a slot for an uninterrupted run.
• Each slot holds one confirmable supplier (Christchurch Toyota), so only orders placed with that supplier can be confirmed. The tools say so, and a capable agent self-corrects.

Then hand it the task

With a connector live, hand the agent the work. Both test prompts live on the experiment pages, where each is explained, so connect here and run them there:

• Build an integration from the docs with Claude Code: the builder test.
• Operate the lifecycle through the Procurement MCP: the operator test.