The Kondado MCP: Connect Your Business Data to ChatGPT, Claude, Codex & Claude Code

Kondado replicates data from 100+ sources to a SQL destination and exposes that destination via the Model Context Protocol (MCP) to leading AI clients. This means you open ChatGPT, Claude, Codex, or Claude Code, ask a question in natural language about your RD Station CRM leads, your Meta Ads spend, or your ERP orders, and the AI responds by reading the data Kondado has already consolidated. No code, no manual exports, no CSV uploads. Learn more about how Kondado connects your data to AI assistants.

Kondado MCP provides a single endpoint (https://mcp.kondado.io/mcp) that connects ChatGPT, Claude, Codex, and Claude Code to your business data via OAuth 2.1 — no API keys to manage. It works because Kondado already consolidated your data: 100+ pipelines replicate to the Kondado Warehouse, and the MCP queries those ready-to-query tables instantly. You can run cross-source analysis impossible elsewhere — ask "which Meta Ads campaigns generated the most paid Bling orders?" and the AI joins data from both sources in one query. Honest limitations include read-only access, maximum 1,000 rows per query, and using KSQL (JSON query language) rather than raw SQL.

What Is the Kondado MCP and How Does It Work?

The Model Context Protocol is an open standard created by Anthropic in 2024 to allow AI clients like ChatGPT and Claude to read data from external sources on demand, without you needing to build a custom pipeline.

The Kondado MCP is the official MCP server for the Kondado platform. It exposes your Kondado Warehouse (the destination where your data has already been replicated) as a queryable source for any MCP-compatible AI client. For a complete overview of how the MCP works with Kondado, see our dedicated MCP guide.

The architecture is straightforward:

1. Your Kondado pipelines (Bling, RD Station, Google Ads, GA4, Omie, Shopify, VTEX, Meta Ads, and 90+ others) extract data from the original APIs.
2. Kondado consolidates everything into SQL tables inside your Kondado Warehouse.
3. The MCP server transforms this warehouse into a source that ChatGPT, Claude, Codex, and Claude Code can query.
4. You ask in natural language; the AI client generates the KSQL query, executes it via MCP, and returns the answer.

Why Should You Use MCP Instead of Calling APIs Directly?

Connecting an AI client directly to the Bling API, GA4, or Meta Ads APIs seems tempting, but fails at three practical points that the Kondado Warehouse solves.

Quota and Rate Limits

Commercial APIs protect traffic with daily and hourly quotas. GA4, for example, releases 200,000 tokens per day per property, shared across all apps querying that account. If you let ChatGPT query GA4 directly, every prompt consumes quota, and your official reports start failing. Kondado replicates once per hour (default frequency) and the AI client queries the SQL destination, which has no quota.

Latency Per Call

When an AI client calls a REST API directly, every question triggers authentication, pagination, and parsing. For slow sources (ERPs like Bling, Omie, Conta Azul), this means seconds of waiting per prompt. Querying the Kondado Warehouse is a SQL query, measured in milliseconds.

Cross-Source Joins

This is the deciding factor. You cannot ask directly "which Meta Ads campaigns generated the most paid orders in Bling last month?" to either API alone. Neither Meta nor Bling knows the answer in isolation. The question requires both datasets to be in the same place. Kondado consolidates everything in the same warehouse, and the MCP allows the AI to cross sources in a single query.

Which AI Clients Can You Connect?

Kondado provides dedicated guides for two major AI families — ChatGPT and Claude — and the MCP server serves all their interfaces with the same endpoint. Detailed setup instructions are available for each client.

ChatGPT, Desktop, and Codex (OpenAI)

The Kondado guide for the OpenAI family covers three interfaces in one article: chatgpt.com (Web), Desktop (Mac/Windows), and Codex CLI. All use the same endpoint https://mcp.kondado.io/mcp and the same OAuth flow. The Web version has been validated on Free/Personal accounts; Desktop inherits data sources configured on the Web. Setup takes under two minutes: add the endpoint, authorize via OAuth, select your Kondado Warehouse.

Claude, Desktop, and Claude Code (Anthropic)

The Kondado guide for the Anthropic family covers three interfaces in one article: claude.ai (Web), Desktop, and Claude Code CLI. Claude Web and Desktop add the MCP at Settings → Connectors → Add custom connector; Claude Code uses the command claude mcp add kondado or a JSON block in settings.json. The OAuth flow discovers configurations automatically from the /mcp endpoint — you only need to paste the URL.

How Do You Configure the Kondado MCP in 3 Steps?

Anyone with a Kondado account connects their first AI client in under 5 minutes. New users complete everything in one session.

1. Connect a data source in Kondado. Choose from 100+ pipelines (Bling, RD Station, Google Ads, GA4, Omie, Shopify, Meta Ads, and others). Kondado authenticates via OAuth, offers replication on the schedule you choose, and directs data to your Kondado Warehouse.

1. Configure the Kondado Warehouse as your destination. The warehouse is a SQL area included in your account. You can also replicate to PostgreSQL, BigQuery, Redshift, Looker Studio, Power BI, and other destinations.

1. Add the MCP server in your AI client. Paste the URL https://mcp.kondado.io/mcp in Settings → Connectors (ChatGPT, Claude) or via CLI command (Claude Code, Codex). OAuth triggers automatically; you authorize the connection; done.

From there, you ask in natural language, in the chat or terminal, and the AI client generates the KSQL query, executes it through the MCP, and returns the answer with data from your account.

What Cross-Source Combinations Can You Ask Through the MCP?

The strength of the Kondado MCP appears when you cross sources that, separately, cannot answer the question. Some practical examples for common profiles.

E-commerce: GA4 + Bling

Ask Claude: "Last month, which campaigns brought traffic in GA4 and generated paid orders in Bling? I want ROAS by campaign." The AI client crosses ga4_traffic_acquisition_daily (sessions and cost by source) with Bling's orders table, returning real ROAS.

Marketing Agency: Meta Ads + Google Ads + RD Station CRM

Ask ChatGPT: "Compare cost per lead from Meta Ads and Google Ads for campaign X last month, and tell me how many of those leads became opportunities in RD CRM." All three sources consolidated in the same warehouse allow the AI to generate an answer with closed numbers.

SMB with ERP: Omie + GA4

Ask Claude Code: "I want to see margin by product category for the last 90 days in Omie, and cross with GA4 visits on the corresponding product pages." The AI assembles the query, returning the table with margin and demand side by side.

Pipeline Analysis: HubSpot or Pipedrive + Meta Ads

Ask Codex: "Which Meta Ads campaigns generated contacts that are now in proposal stage in HubSpot?" The AI client traces the lead journey from ad to open opportunity.

Real ROAS: ERP + Meta Ads

Ask ChatGPT: "Cross my paid orders in Bling last month with spend in Meta Ads — which campaign had the highest ROAS?" The AI joins Bling's paid orders with meta_ads.ad_insights, calculating true ROAS from ERP-confirmed revenue rather than pixel-attributed revenue.

What Are the Current Limitations of the Kondado MCP?

For documentation to be useful, and for you to trust what you are telling your client, manager, or board, it is worth being transparent about the current limitations of MCP version 1.

Read-Only

The Kondado MCP is read-only. No write, update, or delete operations are exposed. This is a protection: the AI client cannot accidentally overwrite your orders table or delete a pipeline.

Maximum 1,000 Rows Per Execution

Each query returns at most 1,000 records, regardless of the limit the AI client requests. The response includes the flag row_cap: 1000 and truncated, so the AI knows the result was cut. For aggregated analyses (sums, averages, top 10), 1,000 rows are usually sufficient; for raw extractions, plan filters before the question.

KSQL Language, Not Raw SQL

The query is generated in KSQL, Kondado's JSON-format query language. Supported AI clients know how to generate KSQL natively, so you do not need to learn the syntax. But it is worth knowing: if you want to audit the query the AI sent, it comes in JSON, not pure SQL.

One Kondado Warehouse Per Session

Each OAuth session is bound to one Kondado Warehouse. If you have multiple warehouses (for example, one per client in an agency), connect each as a separate session in the AI client.

Natural Lag from Original Source

The MCP queries what Kondado has already replicated. If your pipeline with GA4 is configured to run hourly, what the AI sees is data from the last execution, within the natural lag of the original source. GA4, for example, may take up to 48 hours to finalize numbers. This is from the platform, not Kondado.

Who Benefits Most from Using the Kondado MCP?

Three profiles harvest immediate gains with the MCP.

• Marketing teams at agencies: Cross Meta Ads, Google Ads, and CRM in a single prompt, without building a report for each client.
• SMBs with ERPs: Bling, Omie, Conta Azul, and Tiny ERP — queryable in Portuguese or Spanish directly through ChatGPT or Claude, without needing an analyst to open each report.
• Product and data teams: Claude Code and Codex in the terminal allow data questions during the development and analysis workflow itself.

Frequently Asked Questions

What Is the Model Context Protocol (MCP) from Kondado?

The Kondado MCP is an endpoint (https://mcp.kondado.io/mcp) that exposes data from 100+ business sources connected to the platform to AI clients like ChatGPT, Claude, Codex, and Claude Code, via OAuth 2.1.

Do I Need to Know How to Code to Use the Kondado MCP?

No. Setup is visual: you create a pipeline in Kondado, choose the Kondado Warehouse as the destination, and point the AI client to the endpoint. No lines of code are needed.

Which Data Sources Work with the Kondado MCP?

All 100+ Kondado sources work: Google Ads, Meta Ads, VTEX, Shopify, Bling, Omie, Tiny, RD Station, Pipedrive, Google Analytics 4, YouTube, and many others.

Does the Kondado MCP Work on the Free Plan?

Yes, the MCP is available on the 14-day free trial and all paid plans. You only need an active pipeline with the Kondado Warehouse as the destination.

What Is the Difference Between Kondado Warehouse and Other Destinations Like PostgreSQL or BigQuery?

The Kondado Warehouse is a SQL destination managed by Kondado that automatically exposes data via MCP. Other destinations (PostgreSQL, BigQuery, MySQL) receive the data but require their own MCP configuration.

Can I Use the MCP with Multiple Data Sources at Once?

Yes. When multiple pipelines replicate to the same Kondado Warehouse, the AI client can query across all of them in a single question. This is the main advantage — cross-source analysis that no single API can provide.

How Secure Is the MCP Connection?

The MCP uses OAuth 2.1 with Dynamic Client Registration. Your credentials are never stored by the AI client. The connection is read-only, and you can revoke access at any time from your Kondado account settings.

Next Step

Kondado delivers the MCP server at https://mcp.kondado.io/mcp already configured, connects with ChatGPT, Claude, Codex, and Claude Code via OAuth 2.1, and brings your base of 100+ sources into the AI chat you already use. Start with the Kondado free trial, connect your first source, point your AI client to the MCP endpoint, and ask your first question. Support in English and Portuguese via chat on the platform itself.