An MCP server integration in clariBI is a connection to a vendor-hosted tool API. Some tools, the list-shaped endpoints like list_charges or list_invoices, sync into clariBI storage on demand so the AI engine can rank, aggregate, and chart them alongside your other data sources. Other tools, the drill-down and search endpoints like retrieve_balance or search_documentation, stay live and run against the vendor at analysis time. Every tool clariBI exposes is read-only.
How MCP differs from regular data sources
Regular data sources (file uploads, database connections, OAuth integrations like Google Analytics or Jira) ingest data into clariBI on a schedule. The AI then analyses that ingested data. MCP connections work in two modes: materialised for list-shaped endpoints, which sync into clariBI on demand so analytical primitives can aggregate them, and live for drill-down and search endpoints, which each analysis calls in real time. Both modes apply per-tool inside the same connection.
| Regular data source | MCP server | |
|---|---|---|
| Data flow | Vendor → clariBI storage → AI | Synced list tools → clariBI storage → AI; live tools → AI |
| Freshness | Last sync | On-demand sync for list tools, real time for live tools |
| Storage in clariBI | Rows preserved | Synced list-tool rows preserved; live tool results not stored |
| Schema control | clariBI inspects | Vendor publishes tool schemas |
| Best for | Trend analysis, historical reporting | Cross-tool questions, live state, ranked aggregates across tool outputs |
Disconnecting an MCP server cascades to every dashboard, report, and saved analysis that consumed it, the same way disconnecting a regular OAuth integration does. Synced rows are removed alongside the connection.
When to use MCP
Reach for an MCP connection when you want the AI to answer questions against the most current state of a vendor system, or when the vendor exposes data that is awkward to export and ingest. Stripe charges, Jira issues, HubSpot deals, PostHog events, and Notion pages all fit this pattern.
Stick with regular data sources when the vendor does not have an MCP server, when you need long-term trend analysis across snapshots, or when API rate limits make live calls impractical.
Supported services
Every vendor below has a dedicated setup guide linked from its name. Each integration uses a vendor-hosted MCP server (no software to install) and exposes a curated set of read-only tools.
Payments and finance
- Stripe: Payments, customers, subscriptions, invoices, disputes, and payouts from Stripe, plus searchable Stripe documentation.
- Plaid: Aggregated bank transactions, balances, and account data through Plaid, across every financial institution you have linked.
- Ramp: Corporate cards, bills, reimbursements, and department-level spend data from Ramp, for real-time finance reporting.
- PayPal: Transactions and subscription data from your PayPal Business account, alongside Stripe and Square in cross-processor reporting.
- Square: Payments, orders, customers, and location data from Square, for retail and hospitality reporting across stores.
CRM and sales
- HubSpot: Contacts, companies, deals, tickets, engagements, and marketing data from HubSpot CRM and Marketing Hub.
- Attio: Records, custom objects, and workspace structure from Attio CRM, queryable against any schema your team has set up.
Email and marketing
- Klaviyo: Email and SMS campaigns, flows, profiles, events, lists, and segments from Klaviyo, for live email marketing analytics.
Project management and productivity
- Atlassian: Jira issues, Confluence pages and spaces, and Compass components from your Atlassian Cloud, for engineering and docs reporting.
- Linear: Issues, projects, cycles, and team data from Linear, with cycle-level reporting and cross-team velocity built in.
- Asana: Tasks, projects, teams, and workspaces from Asana, including comments and activity for portfolio-wide reporting.
- Notion: Pages, databases, blocks, and workspace content from Notion, with cross-database query support and full-text search.
- monday.com: Boards, items, columns, and users from monday.com, for workflow-state reporting across marketing, ops, and creative teams.
- Airtable: Bases, tables, records, and schema metadata from your Airtable workspaces, queryable across the whole no-code stack.
Customer support
- Intercom: Support conversations, contacts, admin teams, and tags from Intercom, designed for AI-first search across customer messages.
Product analytics
- PostHog: Events, funnels, retention, feature flags, cohorts, and session-replay metadata from PostHog, all read-only via API key.
- Amplitude: Cohorts, behavioral graphs, saved charts, and Session Replay context from Amplitude, for retention and growth questions.
- Mixpanel: Event analytics, funnels, retention, and segments from Mixpanel, queryable in plain language across your tracked events.
Developer and infrastructure
- Sentry: Errors, events, releases, and project health from Sentry, for engineering reliability questions answered in plain language.
- Vercel: Deployments, projects, build logs, and team data from Vercel, for release-correlated reporting on your web apps.
- GitHub: Repositories, pull requests, issues, commits, and code search from GitHub, for engineering velocity and review-cycle reporting.
- Cloudflare: Workers, zones, DNS records, KV namespaces, and Web Analytics from Cloudflare, for traffic and infrastructure reporting.
- Supabase: Projects, table schemas, edge functions, and safe read-only SQL execution against your Supabase Postgres databases.
- Neon: Serverless Postgres projects, branches, and safe read-only SQL execution against any Neon database you have access to.
Content and assets
- Webflow: Sites, CMS collections, and structured content items from Webflow, for editorial dashboards and content auditing.
- Wix: Site metadata, page structure, and published-status data across every Wix site you own.
- Canva: Design metadata, brand kits, folder structure, and creation history from Canva.
- Box: File metadata, folder structure, version history, and content search from Box.
- Egnyte: File metadata, folder structure, and content search from your Egnyte workspace.
- Cloudinary: Image and video asset metadata, transformations, and folder structure from Cloudinary.
Security and data access
Three guarantees apply across every MCP integration:
- The tools the AI engine can call are read-only. No writes, sends, refunds, or destructive operations.
- Tokens and API keys are encrypted at rest and never returned to the browser. The settings page shows redacted values only.
- Disconnect at any time from Settings → Integrations. Disconnecting cascades to every dashboard, report, and saved analysis that consumed the connection.
Getting started
- Open Data Sources from the clariBI sidebar.
- Click Add data source.
- Switch to the MCP Servers tab.
- Pick a vendor and follow the per-vendor setup guide.
- Once connected, ask a question in conversational analytics or build a dashboard widget against the new data source.