Loading data through MCP

An assistant connected to clariBI can add data in three ways. Pick the one that matches where the data is: already in the chat, at a public URL, or in a Google, Meta or Atlassian account.

Choose a path

All three need the data-sources:write scope, which MCP API keys include by default.

Inline upload

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "upload_data_source",
    "arguments": {
      "name": "Q3 sales export",
      "format": "csv",
      "data_base64": "<base64 of the file>",
      "description": "Exported from the CRM on 1 October",
      "wait_seconds": 30
    }
  }
}

Formats: csv, tsv, json, xlsx, xls, txt, pdf. The file is checked and processed like an upload in the web app, and the tool returns a data_source_id. With wait_seconds above 0 (up to 60) the call waits for processing; otherwise poll get_data_source_schema until its status is active.

Plan limits apply

Each upload counts toward the plan's data sources, file uploads and storage. On the Trial that is 3 data sources, 5 file uploads and 1 GB; Starter allows 10 sources, 50 uploads and 5 GB; Professional 50, 500 and 10 GB; Enterprise 100, 5,000 and 100 GB. The error message names the limit that was reached.

Public URL

{
  "jsonrpc": "2.0",
  "id": 2,
  "method": "tools/call",
  "params": {
    "name": "ingest_url_data_source",
    "arguments": {
      "name": "Open data: city budgets",
      "url": "https://example.org/exports/budgets-2026.csv"
    }
  }
}

format is optional; without it clariBI uses the Content-Type header and the file extension. The file is fetched once and stored; it is not fetched again later.

Network rules

The server resolves the address itself and refuses:

• private and internal networks (RFC 1918 ranges, loopback, link-local, IPv6 unique local addresses);
• cloud metadata addresses such as 169.254.169.254;
• anything other than http and https.

Redirects are checked at every hop, with at most 3 redirects.

OAuth handoff

{
  "jsonrpc": "2.0",
  "id": 3,
  "method": "tools/call",
  "params": {
    "name": "request_oauth_integration_url",
    "arguments": { "provider": "google", "integration_type": "ga4" }
  }
}

The response has an authorize_url, a handoff_id and expires_in_seconds. The assistant shows the URL; the user opens it and approves. The assistant then polls:

{
  "jsonrpc": "2.0",
  "id": 4,
  "method": "tools/call",
  "params": { "name": "check_integration_status", "arguments": { "handoff_id": "<id>" } }
}

The status is pending, connected, failed or expired. When connected, the response includes the connection_id and, once the first sync has run, the data_source_id.

After loading

When get_data_source_schema reports active, ask a question with run_analysis:

{
  "jsonrpc": "2.0",
  "id": 5,
  "method": "tools/call",
  "params": {
    "name": "run_analysis",
    "arguments": { "question": "Which channel brought the most revenue in Q3?", "wait_seconds": 30 }
  }
}

What MCP does not do

There is no tool for connecting a database or entering API keys for the API connectors, because that would put credentials into the chat transcript. Connect those from Data Sources in the web app.

Related

• Tool reference
• File uploads
• Data sources