Re:port Flow

Business PDFs — invoices, quotations, delivery notes, statements — generated from reusable templates, by an AI agent, in one turn.

Re:port Flow ships an MCP server so any MCP-capable agent can list a workspace's templates, read a template's parameter schema, and render a filled PDF. No prompt engineering of page layout, no HTML-to-PDF pipeline to babysit: the layout is a template someone designed once, and the agent only supplies the values.

Connect

initialize and tools/list work unauthenticated, so an agent can discover all 10 tools before anyone signs in. Executing a tool needs a Bearer token.

From a Hugging Face agent

from huggingface_hub import MCPClient

async with MCPClient(model="Qwen/Qwen2.5-72B-Instruct") as client:
    await client.add_mcp_server(
        type="http",
        url="https://mcp.re-port-flow.com/mcp",
        headers={"Authorization": f"Bearer {access_token}"},
    )

The Hugging Face SDK has no MCP OAuth flow of its own, so fetch a token once with the helper script in the repository and inject it as a header. The JavaScript client (@huggingface/mcp-client) works the same way — its server config nests under a config key rather than taking flat arguments.

The shortest path to a PDF

list_templates  →  get_design_parameters  →  generate_pdf_sync

Starting from an empty workspace, insert search_gallery_templates → copy_gallery_template to pull a template out of the public gallery first. Copy each template once and reuse the returned designId: every call creates a new design.

Links

• Agent guide — protocol, tools, OAuth, model guidance: https://github.com/re-port-flow/reportflow-mcp/blob/main/agents.md
• Runnable examples — Python, JavaScript, curl: https://github.com/re-port-flow/reportflow-mcp/tree/main/examples
• Developer docs — https://doc.re-port-flow.com
• Source — https://github.com/re-port-flow/reportflow-mcp
• npm — https://www.npmjs.com/package/reportflow-mcp
• Product — https://re-port-flow.com