AI & MCP Access

# Connect to RigForce from your AI assistant.

## Copy the server URL

Add this address in your AI app’s MCP or connector settings.

Server URL

Connection help

If your app offers sign-in, follow its link to RigForce, verify your email and approve the requested permissions.

### Connect with a key

If your app supports custom authorization headers, sign in above, create a key with the permissions you need, and add this header in your app’s connection settings:

Authorization: Bearer YOUR_KEY

Replace YOUR_KEY with your private key. Never put it in a URL, shared prompt or source file.

### Connection protocol

Use https://www.rigforcetech.com/mcp with HTTPS Streamable HTTP. Supported protocol versions: 2026-07-28 and 2025-11-25. OAuth is the main sign-in method; access keys are an alternative for apps that support request headers.

Client setup and verification status

| Client type | Configuration | Original files

| Claude / ChatGPT remote connectors | Add a custom remote MCP connector where your product and account support it. OAuth is required when custom headers are unavailable. Product-specific connection remains unverified. | Do not assume chat attachments reach this service. Use the web upload fallback unless byte transfer has been verified.

| Codex / IDE MCP clients | Add an HTTP MCP server using the address and supported OAuth or Bearer header settings. Exact configuration depends on the client version; not yet verified in these products. | A client with original-file access can use the restricted HTTP upload or small-file tool.

| Protocol integration tests | See the repository's local verification record for the tested SDK and protocol coverage. | File acceptance requires a returned size and SHA-256 checksum, followed by the correct receipt association.

### Tools and permissions

products:read

products_search, products_details, products_match, procurement_info, contact_info

files:write

upload_prepare, upload_inline, upload_cancel and restricted binary upload

inquiries:write

enquiry_prepare, enquiry_summary, enquiry_submit

receipts:read

enquiry_receipt and authenticated downloads of your submitted files

### Query and enquiry workflow

Use a new UUID request_id for each intentional product query. For a transport retry, reuse the same ID and identical arguments. Returned search_id values can be explicitly linked to your enquiry draft. The service records received keywords, filters and the returned public product snapshot, including cached results and failures. It does not receive your complete AI conversation. 

Create an enquiry draft, attach any original files, then call enquiry_summary. Show the complete summary to the customer. Submit with confirm: true and its current confirmation_hash only after the customer confirms. A saved enquiry number confirms receipt, not an order or email delivery.

### Photos and attachments

JPG, PNG, WebP, PDF, Excel, CSV, Word and TXT are supported: 5 files, 10 MB each, 25 MB per enquiry. Call upload_prepare with a stable UUID upload_id, draft ID, file name and actual size. Send original bytes using PUT to its returned URL with your authorization and the returned X-Upload-Token header within 10 minutes. Retries must keep the same file identity and content.

upload_inline accepts Base64 for files up to 1 MB decoded. If your client cannot send the original bytes, open the returned link to this page, sign in and upload to that draft. Refresh the summary in the client afterwards. Unsubmitted files expire after 24 hours. Files are stored privately; receipt means received, not malware-scanned. No OCR or AI document interpretation is provided.

After an interrupted upload, reselect the original file on this page to resume the same upload, including after a page reload. To replace it, explicitly remove the incomplete upload here or call upload_cancel with its draft and upload IDs. Received files remain attached. Review the updated summary again before submitting.

### Usage limits and errors

Each customer shares limits across all keys and clients: 30 tool calls per minute, up to 1,000 query units and 500 returned product records per day. Batch references count separately. Search returns at most 20 records per page and 5 pages; details accept 5 IDs. Uploads allow up to 100 MB per customer per day, including failed transfers. Shared service budgets can reduce availability before these limits.

For quota_exceeded or busy_retry_later, stop automatic loops and retry later. For query_incomplete, keep the request ID and report that completion is unknown. For changed content use a new ID; do not treat service or logging errors as “no products found”. For product_changed_review_required, update the draft with enquiry_prepare and ask the customer to confirm its new summary. Files and successful saves survive notification failures, including exhausted email budgets.

Machine-readable website entry: [llms.txt](https://www.rigforcetech.com/llms.txt). Public pages link to their Markdown version. This aids discovery; it does not automatically install or authorize a connector in every AI client.

[Email: sales@rigforcetech.com](mailto:sales@rigforcetech.com)[WhatsApp: +86 135 1081 6743](https://wa.me/8613510816743)

Source: https://www.rigforcetech.com/en/resources/mcp