Skip to navigation

Paperwork MCP connector

Parse, extract, classify, compose, redact, fill and redline documents from Claude and other MCP clients.
View as Markdown

The Paperwork MCP connector gives an MCP client one tool for each Paperwork task. Each tool calls one or more /v1 routes of the Paperwork API with your credential.

This connector is different from the MCP server at the host root. The root server exposes the full API through two Code Mode tools. The connector exposes fixed tools with a clear read or write type.

URL

The connector lives at https://mcp.cloudraker.com/paperwork. OAuth discovery for this URL is at https://mcp.cloudraker.com/.well-known/oauth-protected-resource/paperwork.

Add the connector in Claude

  1. In Claude, open Settings > Connectors.
  2. Select Add custom connector, or find CloudRaker Paperwork in the connector directory.
  3. Enter the URL https://mcp.cloudraker.com/paperwork for a custom connector.
  4. Select Connect, then sign in with your CloudRaker account.

Authentication

The connector accepts two Bearer credentials:

  • CloudRaker sign-in (OAuth 2.0). Claude runs the sign-in flow. The token is valid only for the /paperwork URL.
  • Organization API key. Other MCP clients can send an organization API key in the Authorization header.

The /v1 API accepts an OAuth sign-in only from an organization administrator. A member who is not an administrator gets 403 on every tool. Use an organization API key for a member account.

claude mcp add --transport http cloudraker-paperwork https://mcp.cloudraker.com/paperwork \
--header "Authorization: Bearer $RAKERONE_API_KEY"

The connector does not keep your credential. Tool results never contain it.

Tools

A read tool only reads data. A write tool creates a file, starts a run or spends credits.

ToolTypeWhat it does
upload_fileWriteAdds a file from a public URL or from base64 content (10 MB maximum). Returns the file object. CloudRaker parses the file and uses credits unless processing is none.
list_filesReadLists the files in the organization API workspace, newest first.
get_fileReadReturns one file with its status and signed links when it is ready.
parse_documentWriteParses a file to Markdown and JSON. Returns links to the output.
extract_dataWriteExtracts structured data from one or more files, with an optional JSON Schema.
classify_documentWriteClassifies one or more files, or their pages, into 2 to 50 classes.
compose_documentWriteRenders a PDF from a saved compose template and a data object.
list_compose_templatesReadLists the saved compose templates of the organization.
redact_documentWriteMakes a redacted copy of a document or recording. The original file does not change.
fill_formWriteFills a PDF form from field values or from source files.
inspect_formReadLists the fillable fields of a PDF form. The field names are the keys that fill_form accepts.
redline_documentWriteApplies a list of tracked edits to a Word document and returns the new file.
get_runReadReturns the status and result of a run.
list_runsReadLists recent runs.

Runs

The parse, extract, classify, redact and fill tools start a run. Each tool waits up to 25 seconds for the result by default. If the run is not done, the tool returns the run id and status. get_run returns the result.

A failed run is not a tool error. The run result has the status failed and an error code.

Errors

A tool error contains the HTTP status, the error code, the message and the request ID. Send the request ID to support when you report a problem. See Errors for the codes.

Credits

Write tools that run a capability use credits, the same as the /v1 routes. A tool returns 402 when the organization has no credits or the plan does not include the capability. Read tools do not use credits.

Privacy and support

The CloudRaker privacy policy applies to all data that the connector reads or writes. For help, send an email to support@cloudraker.com.

Where to go next