# Parseur MCP server

Parseur extracts structured data from documents (PDFs, emails, scans, spreadsheets) and delivers it by webhook or export. This server lets an AI assistant run a Parseur account for its user: create mailboxes (parsers), define the fields to extract, upload documents, read the results, and set up webhooks and exports. Every tool acts on the account of the signed-in user, within the permissions of their role.

Endpoint: https://api.parseur.com/mcp (JSON-RPC over HTTP POST, Streamable HTTP transport, no sessions).

Authentication: OAuth 2.1 with PKCE, or a Parseur API key as bearer token. Discovery starts at https://api.parseur.com/.well-known/oauth-protected-resource/mcp; clients may register dynamically.

## Getting an account

Anyone, including an automated agent, can sign up at https://app.parseur.com/signup with an email address and a password, then confirm that address by following the link in the confirmation email; only confirmed accounts can log in and authorize this server. Signing in with Google on the authorization page needs no email confirmation.

## Example requests

- "Create a mailbox for my supplier invoices and extract the invoice number, date, total and line items."
- "Upload this PDF to my invoices mailbox and show me what was extracted."
- "Send every new parsed invoice from that mailbox to my webhook at https://example.com/hooks/invoices."

## Data handling

Documents and their results are stored in the user's Parseur account. To parse them, Parseur sends their content to its processing providers (OCR and AI engines), listed in the privacy policy. Each tool call is logged with its arguments, cut short, with file contents, email bodies and credentials masked. These tools also reach outside Parseur. upload_from_url downloads the URLs it is given. update_parser_fields and replace_all_parser_fields can add a LINK field: Parseur then downloads the URLs found in each document. Once create_webhook or enable_webhook attaches a webhook, Parseur posts the data of the events it fires on (e.g. each new result, reprocessed ones included) to the URL the user chose. So every tool that processes a document (uploads, copy_document, reprocess_document, split_document, reverse_split_document) can download LINK URLs and post to the enabled webhooks. search_ideas sends its query to the public idea board at feedback.parseur.com (Feature Upvote). submit_idea and upvote_idea post to that board, with the user's name and email, only when the user agrees. send_mcp_feedback sends its message to the Parseur team, only when the user agrees. Last, the export and upload links that tools return work without signing in: anyone who has one can download the results or upload documents.

Privacy policy: https://parseur.com/privacy

## Troubleshooting

- 401 Unauthorized: the access token expired or was revoked; reconnect the server to sign in again.
- A tool is missing from the list: the user's role does not allow it; an account admin can grant the permission.
- A document ends in PARSEDKO or EXPORTKO: get_document_logs tells why.
- Documents end in QUOTAEXC: the plan's credits are used up; get_usage shows the quota.

## Support

support@parseur.com, and the help center at https://help.parseur.com.
