Use Cogny from Claude, ChatGPT or Your Own Agent

Every Cogny workspace is also an MCP server. Any agent that speaks MCP (Claude, ChatGPT, Codex, Cursor, Lovable, the Cogny CLI) can connect to it and use the same tools the in-app agents use: query the warehouse, read and write context, list and update tickets, run reports, and call every connected integration through Cogny's permission model. Cloud customers use this alongside the app; Solo customers use it as the product.

Where

Settings → Data & Integrations → Connect AI Clients. The panel has a tab per client with copy-and-paste instructions. The endpoint is https://app.cogny.com/mcp/w/<your-workspace-id> for clients that take a URL, and https://app.cogny.com/mcp for ChatGPT, where you pick the workspace during authorisation.

Client	How
Claude Code	`claude mcp add --transport http cogny https://app.cogny.com/mcp/w/<workspace-id>` then authorise in the browser
Codex	`codex mcp add cogny --url https://app.cogny.com/mcp/w/<workspace-id>`
Claude.ai	Settings → Connectors → Add custom connector, name it `cogny`, paste the URL, authorise
ChatGPT	Settings → Apps → Advanced settings → Developer mode → Create app, authentication OAuth, URL `https://app.cogny.com/mcp`. It may ask you to authorise twice.
Cursor and other `.mcp.json` clients	The panel shows the JSON block to paste
Lovable	Connectors → Custom, paste the URL
Cogny CLI	`npx @cogny/cli login --api-key <key>`, then `cogny tools list` and `cogny status`

Authorisation is OAuth; no key is needed except for the CLI and .mcp.json, where you generate an API key in the same panel. The key is shown once.

A skill file at cogny.com/SKILL.md teaches an agent how to use the tools well, and llms.txt describes the product for agents that browse it.

What your agent can do

Roughly everything the in-app chat can:

Data: list datasets and tables, inspect schemas, run SQL against your warehouse.
Context: browse, search, read and write context-tree nodes.
Reports and tickets: list reports and read sections, list, create, assign and update tickets, add feedback and comments, link a pull request to a ticket.
Integrations: every connected server's tools, namespaced by integration, under the same auto-approve, ask-first and blocked settings you set in the app.
Connect integrations: list what is available, start an OAuth flow or set a token, without opening the dashboard.
Status: subscription, credits and connection health.

Cogny's own integrations go deep on writes (campaign creation, budgets, keyword lists, page publishing) where most third-party MCP servers stop at reads. That is why the founders built them.

When to use your own agent

You are a specialist who wants execution speed. Twenty years in the Google Ads interface, and you still change budgets faster by asking an agent.
You want to run a ticket on your own tokens. Copy the ticket ID into your agent; the work is still tracked in Cogny, but the inference is paid by your subscription rather than your Cogny credits. Useful for large coding tickets.
You already have an analysis layer and want Cogny's data and execution inside it.
You are checking in from wherever you are. "What happened to the pricing-page experiment?" from your phone.

When not to: if you are not an SEO or growth expert, a general model's ten suggestions include eight you should not run, and there is no context, memory, prioritisation or follow-up behind them. That is what Cloud adds. The founders are direct about this: a non-expert running Solo for SEO can make performance worse.

Credits

Calls from your own agent to a Cloud workspace are metered like in-app calls: a flat minimum per integration call plus token costs where Cogny pays for inference. Your agent's own reasoning is billed by your agent's provider, not by Cogny. On Solo, integration calls are unmetered under fair use.

Team use

Every team member can connect their own client with their own login; roles apply the same way as in the app. A viewer can read; a member can act within the permissions you set.