Ask Cogny's Analyst From Your Own Agent: MCP Tasks Support

Until today, your own agent worked Cogny tool by tool: query Google Ads, read GA4, compare, query again. That works for a quick number. For a question like "why did conversions drop, and what should we change?", it means your agent spends a long session stitching together data Cogny's own analyst already knows how to read.

Now your agent can hand the whole question over.

run_cogny_analysis

run_cogny_analysis takes a question and gives it to the same analyst that answers in Cogny's chat. It works through your connected platforms, analytics, Search Console, your warehouse on Cloud, and the business context Cogny built from your website. It comes back with a written analysis: the findings, the numbers behind them, and concrete recommendations.

• It reads, and changes nothing. The analyst only gets read tools for your connected accounts. Acting on the recommendations stays with you and your agent.
• It takes one to several minutes. The question runs on Cogny's side, so your agent isn't tied up making dozens of calls.
• It's saved in the app. Each analysis is also a chat in Cogny, so you can open it, see every query it ran, and keep asking.

MCP Tasks: a task to follow instead of a call that hangs

A tool call that takes minutes doesn't fit the normal request-and-answer model of MCP. Clients time out, and connections drop.

Cogny now supports the MCP Tasks extension, which Claude is adding support for. When your client supports it, run_cogny_analysis answers straight away with a task. Your agent carries on with other work, checks the task's progress ("Step 4: Run report in Google Analytics 4"), and picks up the answer when it's ready. It can cancel the task at any point.

Loop runs work the same way: run_automation returns a task your agent can wait on, instead of polling for the run's status.

Clients without Tasks support aren't left out. They get a task_id and fetch the answer with get_cogny_analysis.

You confirm before credits are spent

An analysis is charged in credits by usage, like a question in Cogny's chat:

• Cogny AI, Cogny Pro and Cloud: from your monthly credits, then top-ups.
• Solo: from top-up credits.
• Free: from the 50-credit welcome gift.

Pass max_credits to cap a single analysis. It stops at the cap and returns what it has found.

If your client can show forms (MCP elicitation), Cogny asks first: the task opens with a confirmation that names the question, the cap and your current balance, and nothing runs until you say yes. Live loop runs, which can change your connected accounts, ask the same way. With no credits left, the tool says so and links to top-ups instead of starting.

Try it

Connect Cogny to your agent (setup guide) and ask:

Ask Cogny why our paid search conversions dropped over the last 14 days and what we should change. Keep it under 50 credits.

Every tool your agent can reach, including both of these, is listed in the MCP docs.