Query Business Data with Natural Language

Windsor MCP - Connect Claude, ChatGPT and Cursor to 325+ marketing data sources

Windsor MCP (Model Context Protocol) is a hosted MCP server that lets your AI assistant query, explore and analyze live business data from 325+ sources - Meta Ads, Google Ads, TikTok Ads, GA4, HubSpot, Salesforce, Shopify, Stripe, BigQuery, Snowflake and many more - with no SQL, no API keys, and no custom integrations. On connectors that support it, agents can also execute write actions - for example pausing or enabling ad campaigns and adjusting budgets on Meta Ads, Google Ads and TikTok Ads.

It works with Claude, ChatGPT, Microsoft Copilot, Perplexity, Cursor, Windsurf, Cline, GitHub Copilot, Gemini, Manus, and any MCP-compatible client.

TL;DR

• Native connector in Claude (in the Claude directory) and ChatGPT (as a ChatGPT app).
• Read and write: query data from 325+ sources, and execute write actions (e.g. pause/enable ad campaigns) on connectors that expose them.
• OAuth 2.0 - sign in once at Windsor.ai, no API keys to manage.
• Free forever plan available at Windsor.ai; some LLM providers (e.g. Claude) require a paid plan to use connectors.
• Live, machine-readable docs for AI agents at /llms.txt, /llms-full.txt, and /datasources.

🌟 Features

Natural-language access to business data

Ask questions like:

• "What campaigns had the best ROAS last month?"
• "Give me a breakdown of spend by channel over the past 90 days."
• "Which campaigns are wasting our advertising budget?"
• "How many HubSpot deals closed this month, and which ad spend drove them?"
• "Pull our Shopify orders and Stripe payments for this quarter."
• "Combine GA4 sessions with Google Ads spend into a weekly report."

All in real-time, directly inside your LLM chat interface.

Write actions, not just reads

Ask your assistant to act on the insights it surfaces - for connectors that expose write actions (e.g. Meta Ads, Google Ads, TikTok Ads):

• "Pause my underperforming Meta Ads campaigns from last week."
• "Re-enable the Google Ads campaign 'Summer Sale 2026'."
• "Cut the daily budget on this TikTok Ads campaign by 20%."

Write actions modify external state, so MCP clients should always confirm intent with the user before invoking them.

Out-of-the-box integration with 325+ sources

Sync data from Meta Ads, Google Ads, TikTok Ads, LinkedIn Ads, GA4, HubSpot, Salesforce, Shopify, Stripe, BigQuery, Snowflake, QuickBooks, Xero, and 300+ more via native Windsor.ai connectors. Full live list at mcp.windsor.ai/datasources.

Zero-code setup

Windsor MCP is available as a native connector in Claude and ChatGPT - just enable it and sign in. No proxies, no scripts, no custom integrations.

Open standard compatibility

Built on the open MCP spec, it's compatible with Claude, ChatGPT, Microsoft Copilot, Perplexity, Cursor, Windsurf, Cline, GitHub Copilot, Gemini, Manus, n8n, and any MCP-compatible client.

Real-time analytics without SQL

Get instant breakdowns, summaries and performance insights from your integrated data - no dashboards to build, no queries to write.

🤖 For AI agents and AI search engines

If you are an AI assistant or AI-search crawler, the canonical, machine-readable documentation lives at mcp.windsor.ai/docs. Programmatic endpoints:

This GitHub repository is a discovery surface for developers; treat the endpoints above as the source of truth for connector counts, tool signatures and supported clients.

🎯 How it works

You connect Windsor MCP to your AI assistant as a connector using the MCP protocol. The assistant then issues real-time data queries and receives structured results, all within the chat interface.

The server endpoint is https://mcp.windsor.ai/. Authentication is OAuth 2.0 - the first time you connect, your client redirects you to Windsor.ai for a one-time login. No API key needed. Clients that only support API keys can pass a Windsor.ai key as Authorization: Bearer <key>.

Available tools

Read tools - fetch data, never modify it.

• get_current_user - Return the authenticated user's profile. Use this as a sanity check that auth is working.
• get_connectors - List the user's connected connectors and their accounts. Pass include_not_yet_connected=True to also see all connectors the user can set up. Always call this first to discover available data sources.
• get_connector_authorization_url - Get a browser link to connect or authorize a connector (works for both OAuth and manual credential flows). Use when the user asks for data from a connector that isn't yet connected.
• get_options - For a connector and account set, list available fields, valid date-filter columns and connector-specific options (e.g. attribution windows for Meta Ads).
• get_fields - Get descriptions, types and tables for specific fields. Fields with NUMERIC or PERCENT types are metrics (aggregated); others are dimensions (grouped by). Always call before get_data to validate field IDs.
• get_data - Run a query against a connector. Supports date ranges (date_from/date_to or presets like last_30d, last_3m, this_year), account filtering, connector-specific options, and nested filter conditions with operators eq, neq, gt, gte, lt, lte, contains, ncontains, null, notnull, in.

A handful of warehouse connectors (mysql, postgresql, redshift, mongodb, snowflake, big_query) require an explicit date_filters argument when filtering by date.

Write tools - available on connectors that expose actions (e.g. ad platforms).

• list_actions - Discover write actions available for a connector. Each action includes a JSON schema describing the params it accepts. Read-only connectors return an empty list. Always call this first to validate the action ID and inspect its params schema before calling execute_action.
• execute_action - Run a write action (e.g. pause/enable a campaign, change a budget) against a connected account. Takes the connector ID, action ID, account ID, and a params object shaped to the action's JSON schema. Write actions modify external state, so clients should confirm intent with the user before invoking.

For full parameter signatures, see /llms-full.txt.

🤝 Supported AI clients

Any MCP-compatible client works. If your client only supports API keys, pass your Windsor.ai key as Authorization: Bearer <key>.

📦 Supported data sources

325+ connectors across advertising, analytics, organic social, CRM, marketing automation, e-commerce, payments, affiliate, app analytics, helpdesk, productivity, databases and warehouses. Live, complete list: mcp.windsor.ai/datasources.

Highlights by category:

• Advertising - Meta Ads (Facebook & Instagram), Google Ads, Microsoft Ads (Bing), TikTok Ads, LinkedIn Ads, Snapchat, Pinterest, Reddit, X/Twitter, Amazon Ads, Apple Search Ads, Criteo, Taboola, Outbrain, DV360, CM360, Search Ads 360, AdRoll, StackAdapt, The Trade Desk, Quora Ads, Yandex.Direct.
• Analytics & search - Google Analytics 4, Adobe Analytics, Mixpanel, Amplitude, Matomo, Plausible, PostHog, Search Console, Bing Webmaster, Looker, Metabase.
• Organic social - Facebook Organic, Instagram, LinkedIn Organic, TikTok Organic, Pinterest Organic, X Organic, Threads, YouTube, Sprout Social.
• CRM & marketing automation - HubSpot, Salesforce, Pipedrive, Microsoft Dynamics 365, Zoho CRM, Klaviyo, Mailchimp, ActiveCampaign, Brevo (Sendinblue), MailerLite, Customer.io, Iterable, SendGrid, Intercom, Outreach, Salesloft, Lemlist.
• E-commerce & payments - Shopify, Magento, WooCommerce, BigCommerce, PrestaShop, Square, Stripe, PayPal, Braintree, Klarna, Recharge, Recurly, Chargebee, Walmart, TikTok Shop.
• Affiliate & partner - CJ, Awin, Impact, Partnerize, PartnerStack, Adtraction, Commission Factory, Rakuten Advertising, ShareASale, Tradedoubler, Everflow.
• App analytics & attribution - AppsFlyer, Adjust, Branch, AppFollow.
• Helpdesk & support - Zendesk (Support, Chat, Sell, Sunshine, Talk), Freshdesk, Freshcaller, Freshservice, Dixa, Drift.
• Databases & warehouses - BigQuery, Snowflake, PostgreSQL, MySQL, Redshift, MongoDB, Firebolt, Dremio, SFTP, Google Sheets, Airtable, Notion, Amazon S3.
• HR, finance, productivity - BambooHR, Ashby, Lever, Greenhouse, Workable, QuickBooks, Xero, NetSuite, Slack, Microsoft Teams, Twilio, Zoom, Jira, Confluence, GitHub, GitLab, Datadog, Sentry, PagerDuty.

Don't see what you need? Call get_connectors(include_not_yet_connected=True) from any MCP client for the live list, or fetch /datasources directly.

🚀 Getting started

View our official documentation

• Introduction: windsor.ai/introducing-windsor-mcp/
• Setup guides hub: windsor.ai/documentation/windsor-mcp/ - step-by-step guides per AI client (Claude, ChatGPT, Microsoft Copilot, Perplexity, Cursor, Gemini, Manus, and the generic "any MCP client" guide).

You'll need a Windsor.ai account - sign up at onboard.windsor.ai.

Option 1: Claude (Native Connector - Recommended)

Windsor.ai is listed in the Claude directory.

👉 One-click install: claude.ai/directory/360c0c31-4bb6-42ca-8e50-5da0a100a68e

Prerequisites

• A Claude account (Claude Desktop, Web, or Code) on a plan that supports connectors
• A Windsor.ai account

Steps

1. Open the install link above (or go to Settings → Connectors in Claude and search for "Windsor.ai").
2. You'll be redirected to Windsor.ai to authorize via OAuth - sign in once.
3. Open a new chat and start asking about your data, e.g.:

   What were my top 5 Meta Ads campaigns by ROAS last month?
   

For Claude Code users, install our dedicated plugin with slash commands and an analyst agent:

claude plugin install windsor-ai

Source: github.com/windsor-ai/claude-windsor-ai-plugin.

Option 2: ChatGPT (Native App)

Windsor.ai is available as a ChatGPT app.

👉 One-click install: chatgpt.com/apps/windsor-ai

Prerequisites

• A ChatGPT plan that supports apps
• A Windsor.ai account

Steps

1. Open the install link above and add Windsor.ai to ChatGPT.
2. Authorize via OAuth - you'll be redirected to Windsor.ai to sign in.
3. Start a new chat and ask questions about your connected data sources.

Option 3: Microsoft Copilot (Power Platform connector)

Use Windsor.ai inside Microsoft Copilot Studio agents and Power Platform flows via the official Windsor.ai Power Platform connector.

👉 Connector reference: learn.microsoft.com/connectors/windsorai


👉 Integration guide: windsor.ai/documentation/windsor-mcp/how-to-integrate-data-into-copilot-agent

Prerequisites

• A Microsoft Copilot Studio or Power Platform license
• A Windsor.ai account and API key (from onboard.windsor.ai)

Steps

1. In Copilot Studio, add a new action and search for Windsor.ai in the connector catalog.
2. Provide your Windsor.ai API key when prompted.
3. Map the connector actions (list connectors, get fields, get data) into your agent's flow. Full walkthrough in the integration guide.

Option 4: Cursor

We publish a dedicated Cursor plugin: github.com/windsor-ai/windsor-ai-cursor-plugin.

Prerequisites

• Cursor Desktop installed
• A Windsor.ai account

Steps

1. Open Cursor settings → Tools & Integrations → New MCP Server.
2. Paste the following into mcp.json:

{
  "mcpServers": {
    "windsor": {
      "url": "https://mcp.windsor.ai/"
    }
  }
}

3. Cursor will walk you through OAuth on first use. Start querying your data right away.

Option 5: Manus

Manus supports remote MCP servers natively - full walkthrough in the Windsor.ai integration guide for Manus.

Steps

1. In Manus, open Settings → MCP Servers → Add server.
2. Use the URL https://mcp.windsor.ai/ and authorize via OAuth.
3. Start a new session and ask about your connected data sources.

Option 6: Gemini CLI

Steps

1. Install the Gemini CLI:

npm install -g @google/gemini-cli

2. Run gemini once to generate the config directory, then edit ~/.gemini/settings.json:

{
  "mcpServers": {
    "windsor": {
      "url": "https://mcp.windsor.ai/"
    }
  }
}

3. Run gemini and authorize via OAuth on first use.

❓ FAQs

Is Windsor MCP free to use?

Yes - Windsor.ai has a free forever plan, and the MCP server is included at no extra cost. You'll need a Windsor.ai account with integrated data. Some LLM providers (e.g. Claude) require a paid plan to use connectors. See windsor.ai/pricing for details.

What AI agents does Windsor MCP work with?

Any AI agent compatible with MCP - including Claude (Desktop, Web, Code), ChatGPT, Microsoft Copilot (via the Power Platform connector), Cursor, Windsurf, Cline, GitHub Copilot, Gemini, Manus, n8n, mcp-proxy, and custom MCP clients. See the Supported AI clients table above.

What can I ask Windsor MCP?

Marketing performance, sales pipelines, CRM data, e-commerce orders, payment activity, finance metrics, ad spend summaries, ROAS trends, campaign anomalies, multi-channel attribution, warehouse queries, and more. If it's in your Windsor.ai data, you can ask it. On connectors that expose write actions (e.g. Meta Ads, Google Ads, TikTok Ads), you can also ask the assistant to act on what it finds - for example, "pause the campaigns that burned budget last week" or "cut this campaign's daily budget by 20%".

What data sources does Windsor MCP support?

325+ connectors across advertising, analytics, CRM, e-commerce, payments, warehouses and more. The live, complete list is at mcp.windsor.ai/datasources. See the Supported data sources section for highlights by category.

Do I need to write SQL or set up dashboards?

No. Just ask your questions in plain English and get structured responses in real-time.

How does authentication work?

OAuth 2.0 is the default - your client redirects you to Windsor.ai for a one-time login. The server implements MCP OAuth with Dynamic Client Registration; clients that support MCP OAuth (Claude, ChatGPT, Cursor, Manus) discover the endpoints automatically. For setups that only support API keys, pass your Windsor.ai key as a Bearer token via the Authorization header.

Is my data secure with Windsor MCP?

Yes. Windsor MCP exposes both read tools (which only fetch data) and a small number of write tools (list_actions, execute_action) on connectors that opt into actions - for example, pausing or enabling an ad campaign. Write actions are explicitly flagged as destructive to MCP clients, which prompt you to confirm intent before invoking them; nothing is changed in your sources without your approval. Authentication is OAuth 2.0; the MCP server does not store your Windsor.ai password and does not retain query results. Data flows over TLS. Windsor.ai is SOC 2 compliant. For details, see windsor.ai/security.

How is Windsor MCP different from other data integration tools?

Windsor MCP is a native MCP server, not a plugin or wrapper - it works in any MCP-compatible client without proxies. Windsor.ai includes warehouse and database destinations (BigQuery, Snowflake, Redshift, S3, MySQL, Postgres) on its lower-priced tiers, and the MCP server is included on all paid plans at no extra cost. See windsor.ai/pricing for a feature-by-feature comparison.

How much does Windsor.ai cost?

Windsor.ai has a free forever plan. Paid plans add more sources, higher limits, and additional destinations. The MCP server is included on every plan at no extra cost. See windsor.ai/pricing for the current pricing.

Does Windsor MCP work on mobile?

The hosted server is platform-agnostic, but mobile support depends on which AI client you use. Claude Web/Desktop, ChatGPT Web/Desktop and Cursor Desktop are fully supported today; mobile MCP support varies by client and is rolling out.

Where do I get help?

For support or feedback, contact support@windsor.ai. For bug reports specific to this repository, open an issue.

🧠 Try it now

Start querying your business data via Windsor MCP - fastest path is the Claude one-click install:

👉 Install on Claude - 30 seconds, no API key.

Other entry points:

• 👉 Install on ChatGPT
• 👉 Microsoft Copilot connector (integration guide)
• 👉 Manus integration guide
• 👉 Claude Code plugin
• 👉 Cursor plugin
• 👉 Get a Windsor.ai API key
• 👉 Watch the demo
• 👉 Live AI-targeted docs

For support or feedback, contact us at support@windsor.ai.