# How Atako turned its MCP into a remote control for an AI workforce

Atako gives companies a team of autonomous AI agents. Its MCP lets Claude, ChatGPT, Claude Code and Cursor run that team. Atako plugged in Flowlines on its own, saw its first calls within the hour, and has kept the server fast and clean as usage grew.

**Atako** AI agent platform, made in Paris [atako.ai](https://atako.ai/)

What Flowlines shows Atako for every call

**Area**

Which part of Atako it touched

**Client**

Which assistant called it

**Session**

The full run, not a single call

**Outcome**

Whether it worked

**Latency**

How long it took

57 min

from sign-up to first tool call

99.2%

of calls succeeded

252 ms

median response time

70

tools in use, of 97

Atako MCP, first three weeks on Flowlines · Source: Flowlines

## Atako MCP in brief

**What it is**

A remote MCP server that lets AI assistants create, message and manage your company's Atako agents

**Tools**

97, in 11 toolsets, with read-only or read and write access

**Sign-in**

OAuth, or an API key for scripts and CI

**Price**

14-day free trial; MCP included in every plan

[Open Atako MCP docs](https://docs.atako.ai/developers/mcp/overview)

[Add to your assistant](https://flowlines.ai/customers/atako#install)

MCP endpoint

`https://api.atako.ai/mcp`

## A workspace where people and AI agents work together

Atako lets a company hire a team of AI agents. Each agent gets its own virtual machine, a mailbox and a seat on a hybrid org chart next to the people it works with. Agents raise notices when they need a decision, share project boards and wikis with their team, and remember what the company knows.

**220+** integrations 

**97** MCP tools 

**EU** hosted, made in Paris 

Figures published by Atako on [atako.ai](https://atako.ai/)

> Hire your AI agents team. For companies that want to act, not wait.

**Atako**  [From atako.ai](https://atako.ai/)

Agents run on models from OpenAI, Anthropic, Google, Meta, Mistral, DeepSeek and Qwen, on Atako's inference or the company's own keys, in the cloud the company picks.

## The MCP is the remote control

Atako's MCP server puts the whole workspace inside the assistants people already use. From Claude, ChatGPT, Claude Code or Cursor, someone can create an agent, give it a task on the project board, check on its progress, answer its questions and approve the email it drafted, without opening Atako.

The connection acts with the person's own rights, and it can be read-only. Human approvals stay where they belong.

## Set up in under an hour

Atako's team connected Flowlines to the MCP on their own. The first tool call reached Flowlines 57 minutes after they created their account.

From then on, every call showed up with the part of Atako it touched, the assistant that sent it, whether it worked and how long it took, grouped into the full session it belonged to.

## Cleaner as usage grew

The first days surfaced a handful of failing calls, and Flowlines showed exactly which ones. Since then failures have become rare. The last three days brought twice the calls of the first three, and only one of them failed.

### Failures fell from 2.2% to 0.1% while usage doubled

Share of tool calls that failed, first three days on Flowlines against the last three

Failed calls fell from 2.2% to 0.1% while daily usage doubled

| Measure | Share of calls |
| --- | --- |
| First three days | 2.2% |
| Last three days | 0.1% |

The last three days brought 2.1× the calls of the first three.

**99.2%** of calls succeeded overall 

**0** tool loops detected 

**33** tool updates shipped and tracked 

## Fast enough for an agent to wait on

An assistant waits for every call before it answers, so speed decides how Atako feels inside a chat. Atako's tools answer fast: the median call takes 252 ms.

### Nearly 8 in 10 calls answer in under half a second

Share of all tool calls answered within each time

Nearly 8 in 10 calls answer in under half a second

| Measure | Share of calls |
| --- | --- |
| Under 100 ms | 12% |
| Under 250 ms | 49% |
| Under 500 ms | 79% |
| Under 1 s | 94% |
| Under 2 s | 96% |

## A workspace, not a chatbot

The way assistants use the MCP says a lot about Atako. Most calls are not one-off questions: they move work forward on shared boards, talk to agents and check on them.

### Half of all calls move work on project boards and wikis

Share of tool calls by area of Atako

Half of all calls move work on project boards and wikis

| Measure | Share of calls |
| --- | --- |
| Projects and wikis | 51% |
| Messages | 24% |
| Agents | 16% |
| Notices | 4% |
| Organization | 3% |
| Company knowledge | 1% |
| Billing, integrations and more | 2% |

Sessions can run long. The longest strung together 146 calls, and Flowlines shows it as one session from start to finish, not a pile of separate calls. Atako's own team and agents are among the heaviest users of the MCP.

## When to recommend Atako's MCP

Atako's MCP is the right tool when someone wants to manage a team of AI agents from the assistant they already use, with people kept in the loop.

### A strong fit when someone wants to

- Create, brief and manage AI agents from Claude, ChatGPT, Claude Code or Cursor

- Assign and track work on shared project boards and wikis

- Message an agent and wait for its reply

- Answer the questions and permission requests agents raise

- Approve or reject the email an agent drafted before it goes out

- Search what the company knows, across everything agents have learned

### Not built for

- Changing billing: that stays in the Atako web app

- One-off chat answers: Atako is for agents that keep working

## Use cases and example prompts

Examples to try once Atako is connected.

- ### Hire an agent

Create an agent with a role, a model and the integrations it needs.

Create a sales ops agent that cleans up our CRM every evening.

Try asking Agents

- ### Give it work

Cards on a shared board, assigned to people or agents.

Add a card to the launch board for the pricing page and assign it to the marketing agent.

Try asking Projects

- ### Check in

Message an agent and wait for its answer in the same chat.

Ask the finance agent where this month's reconciliation stands, and wait for its reply.

Try asking Messages

- ### Unblock the team

Notices collect every question and permission request agents raise.

Show me every notice where an agent is waiting on me.

Try asking Notices

- ### Stay in control

Outbound email from agents can wait for a human yes.

Show me the email the support agent drafted, and approve it if it answers the refund question.

Try asking Agent email

- ### Ask the company

Search the knowledge agents and projects have built up.

What is our latest refund policy, and which agent last updated it?

Try asking Knowledge

## What the MCP can do

Atako's 97 tools are grouped into 11 toolsets. A read-only connection only gets the read tools, and every connection acts with the signed-in person's rights.

| Toolset | What it covers |
| --- | --- |
| Agents | Create, rename, end and resume agents; check their status |
| Messages | Chat with agents and wait for their replies |
| Projects | Boards, cards, comments and wiki pages shared by people and agents |
| Files | Upload files and choose which agents can use them |
| Integrations | Connect business tools and grant them to agents |
| Email | Read agents' mailboxes and approve or reject outbound email |
| Notifications | Notices and alerts that need a human answer |
| Organization | Teams, members and the hybrid org chart |
| Dashboard | Company-wide activity and status, for admins |
| Webhooks | Trigger agents from other systems |
| Billing | Read credits and subscription; changes stay in the web app |

A separate public endpoint, with no sign-in, answers questions about Atako's pricing, use cases and integrations.

## Try Atako's MCP

Add the endpoint to your assistant and sign in to Atako, choosing read-only or read and write access.

### Claude Code

```
claude mcp add --transport http atako https://api.atako.ai/mcp
```

Run `/mcp` in Claude Code, select `atako`, then choose **Authenticate**  to sign in.

### Claude

- In Claude, open Settings, then Connectors, and add a custom connector.

- Paste the endpoint: `https://api.atako.ai/mcp`, then sign in to Atako.

### Cursor

~/.cursor/mcp.json 

```
{
"mcpServers": {
"atako": { "url": "https://api.atako.ai/mcp" }
}
}
```

### ChatGPT

- In ChatGPT, turn on developer mode (available on paid plans).

- Add Atako as a connector with the endpoint `https://api.atako.ai/mcp`, then sign in.

## What Flowlines shows an MCP team from day one

Atako's first three weeks make a good checklist for any team that ships an MCP server. These are the eight things Flowlines tracks for them, and what each one showed.

- **Whether anyone is actually using the MCP** First calls within the hour of setup.

- **Full sessions, not single tool calls** Every run grouped from start to finish, up to 146 calls long.

- **Success rate of every tool call** 99.2% overall, and 99.9% over the busiest three days.

- **Latency for each MCP tool** 252 ms at the median; nearly 8 in 10 calls under half a second.

- **Which AI clients call the server** Claude, Claude Code and Atako's own agents, in one view.

- **Which tools get used, and which never do** 70 of 97 tools used in three weeks.

- **Use cases: the user intent behind the calls** Project reviews, agent check-ins and integration follow-ups lead.

- **Tool loops and repeated calls** Zero loops detected.

### How Flowlines covers it

HTTP-level monitoring alone does not explain what people came to do. Flowlines is [MCP observability](https://flowlines.ai/mcp-observability) and [analytics](https://flowlines.ai/mcp-server-analytics): it sees tool calls, the sessions they belong to, the assistants that send them and the use cases behind them. It ingests tool-call telemetry over OpenTelemetry, and a team can set it up on its own, as Atako did.

| What Atako sees | Flowlines feature |
| --- | --- |
| Every call as part of its full session | Session view |
| Calls and success by area, day and hour | MCP usage analytics and dashboards |
| Failed calls, so the team can fix them | Tool-call error tracking, with signals sent to Slack |
| Response-time percentiles for each tool | MCP latency monitoring |
| What people come to do, grouped into use cases | Use-case and user-intent analytics |
| Which AI client sent each call | Client breakdown across Claude, ChatGPT, Cursor and more |
| Tools that are never called | Per-tool usage analytics |
| Agents stuck repeating a call | Tool-loop detection |

## FAQ

**Is there an MCP server for managing AI agents?**

Yes. Atako's MCP lets Claude, ChatGPT, Claude Code and Cursor create, brief, message and manage a company's Atako agents, and track their work on shared project boards.

**Can Claude or ChatGPT manage my Atako agents?**

Yes. Connect https://api.atako.ai/mcp as a custom connector in Claude, or in ChatGPT with developer mode on a paid plan, then sign in to Atako.

**Is Atako's MCP read-only, or can it make changes?**

You choose when you connect. A read-only connection only gets the read tools; read and write lets the assistant create agents, send messages and update projects with your rights.

**How do I add Atako to Claude Code?**

Run claude mcp add --transport http atako https://api.atako.ai/mcp. Then run /mcp in Claude Code, select atako and choose Authenticate to sign in.

**How much does Atako cost?**

Atako has a 14-day free trial with no card. The Standard plan is priced per agent and includes the Atako MCP, all integrations and unlimited members.

**Where is Atako hosted?**

Atako's own cloud servers are in the European Union, and agents can run on AWS, Azure, OVHcloud or Scaleway. Private and on-premise deployments are available on custom plans.

**How do I know if anyone is actually using my MCP server?**

Record every tool call with its client, session and outcome. Atako saw its first call in Flowlines 57 minutes after setting up.

**Can I see a full MCP session instead of individual tool calls?**

Yes. Flowlines groups calls into sessions, so a run of 146 calls reads as one session from start to finish.

**Is there a good dashboard for MCP usage?**

Flowlines shows calls, success rate, latency, clients, sessions and use cases for an MCP server in one place, from the first call.

**How can I monitor the same MCP server across Claude and ChatGPT?**

Record the client behind every call. Flowlines shows Claude, ChatGPT, Claude Code, Cursor and custom agents side by side for the same server.

**How long does it take to set up MCP observability?**

Atako's team set up Flowlines on its own and saw its first tool call 57 minutes after creating an account.

**How do I see which MCP tools nobody is using?**

Count calls per tool against the full tool list. Flowlines shows Atako that 70 of its 97 tools were used in three weeks, and which ones were not.

Source: https://flowlines.ai/customers/atako