clariBI MCP Server

Public reference for the clariBI Model Context Protocol server, a hosted MCP endpoint that lets LLM clients sign users up, upload data, connect cloud sources via OAuth, run natural-language analyses, and generate reports.

This repo holds the public manifest, examples, and reference documentation. The clariBI platform itself is a commercial hosted service; the hosted MCP endpoint is documented here so any spec-compliant MCP client can connect.

Endpoint

Transport	URL	Protocol version
Streamable HTTP	https://claribi.com/mcp/v1/	2025-03-26
SSE (legacy)	https://claribi.com/mcp/v1/sse	2024-11-05

Authentication

OAuth 2.1 with Dynamic Client Registration (RFC 7591). Discovery endpoints:

• Authorization Server Metadata (RFC 8414): https://claribi.com/mcp/v1/.well-known/oauth-authorization-server
• Protected Resource Metadata (RFC 9728): https://claribi.com/mcp/v1/.well-known/oauth-protected-resource

Public clients use PKCE S256. Confidential clients receive a client secret on registration.

Alternative for single-user installations: bearer token Authorization: Bearer claribi_mcp_<key>. Mint a key inside the clariBI web app under Settings > Developer > MCP Server.

Surface

19 tools, 3 resources, 2 prompts. Tool names and JSON schemas are published live at the tool catalog endpoint:

curl https://claribi.com/mcp/v1/  # returns server metadata
curl https://claribi.com/api/mcp-server/tool-catalog/  # full catalog

Tools include register_account, verify_email, check_pricing (all unauthenticated for sign-up from chat), upload_data_source, ingest_url_data_source, request_oauth_integration_url, run_analysis, generate_report, get_usage, get_billing_status, create_checkout_session, and dashboard/report list/get tools.

Official SDKs

Both SDKs are MIT-licensed and codegen-driven from the live tool catalog so the typed method surface never drifts from the server.

Python

pip install claribi-mcp

from claribi_mcp import Client

with Client(api_key="claribi_mcp_...") as client:
    client.initialize()
    result = client.run_analysis(
        question="What was revenue by region last quarter?",
        wait_seconds=30,
    )
    print(result.text)

Source on PyPI: https://pypi.org/project/claribi-mcp/

TypeScript

npm install @claribicom/mcp

import { Client } from '@claribicom/mcp';

const client = new Client({ apiKey: 'claribi_mcp_...' });
await client.initialize();
const r = await client.callTool('run_analysis', {
  question: 'What was revenue by region last quarter?',
  wait_seconds: 30,
});
console.log(r.text());

Source on npm: https://www.npmjs.com/package/@claribicom/mcp

MCP Server Registry

Listed in the official Model Context Protocol Server Registry under the namespace com.claribi/mcp-server:

curl 'https://registry.modelcontextprotocol.io/v0.1/servers?search=com.claribi'

Manifest: server.json.

Pricing and tier gating

MCP Server access is gated by the mcp_server_access feature flag on the clariBI subscription. It ships with Trial (free, 14 days), Starter ($99/mo), Professional ($199/mo), and Enterprise ($999/mo). Excluded from Free and Lite. Tool error responses indicate which quota was hit.

Sign-up tools (register_account, verify_email, check_pricing) do not require a subscription, so a new user can provision a Trial workspace from inside their LLM client without any prior account.

Documentation

• Overview: https://claribi.com/docs/mcp/overview
• Connecting from Claude Desktop: https://claribi.com/knowledge-base/mcp-connect-claude-desktop/
• Connecting from Cursor: https://claribi.com/knowledge-base/mcp-connect-cursor/
• Sign-up via LLM walkthrough: https://claribi.com/knowledge-base/mcp-signup-via-llm/
• API reference: https://claribi.com/knowledge-base/mcp-api-reference/

Security

Vulnerability disclosure: [email protected]. See SECURITY.md.

License

This repository (README, manifest, examples, SDK code) is MIT.

The hosted clariBI platform that the server fronts is a commercial service; using the MCP server endpoint requires a clariBI account.