Adako: ad ops for AI assistants

アダコ · adako.ai

Adako is an MCP server and REST API that connects Claude, ChatGPT, Cursor and other assistants to Google Ads, Meta Ads, ChatGPT Ads, TikTok Ads and LinkedIn Ads.

https://adako.ai/mcp

Ask "how did my campaigns do last month?" and get an answer from the live account, in the account's own currency. Ask "raise that budget to 40 a day" and get a proposal you approve before anything moves. Created campaigns are always paused.

This repository holds the public pieces: the Agent Skills, per-client configuration, one-file installers, a Claude Code plugin, and the MCP Registry entry. The server itself is hosted.

What it can do

233 tools across five platforms, plus Adako's own.

Your client does not see 233 entries. It sees 16 tools callable by name plus 13 routers. Each platform has a read router (google_ads, meta_ads, chatgpt_ads, tiktok_ads, linkedin_ads) and a write router beside it (google_ads_write, meta_ads_write, chatgpt_ads_write, tiktok_ads_write, linkedin_ads_write). Monitors have monitoring and monitoring_write, and diagnostics only reads. A read looks like google_ads(action="execute", tool_name="google_list_campaigns", arguments={...}), a change like google_ads_write(action="execute", tool_name="google_pause_campaign", arguments={...}).

Looking things up never changes an account, so an assistant that asks permission per tool can allow reads once. Every change runs through a _write tool, which asks each time and creates a proposal. search_tools finds the right tool from your own wording and returns its exact call line, and it is free. See adako.ai/docs/routers.

LinkedIn Ads reads are open to everyone: connect the member who administers the ad account and every report, structure and forecast tool works. A change needs that ad account added to Adako's LinkedIn application first — send the nine-digit id to support@adako.ai. The TikTok Ads tools are on the surface, but TikTok accounts cannot be connected until TikTok approves Adako's developer app.

The safety model

Adako is built so that an assistant can be useful without being dangerous.

Details: adako.ai/docs/safety.

Install — five minutes

Two ways to authenticate:

• OAuth — the client opens a browser, you sign in to Adako once. Preferred everywhere it works.
• API key — create one at adako.ai/keys, send it as Authorization: Bearer ak_live_your_key, where ak_live_your_key is replaced by the whole key you copied, prefix included and nothing around it. Use it for clients that read a static config file.

Pick your client.

Claude.ai (custom connector)

Pro and Max: Settings → Connectors → Add custom connector. Paste the URL, click Add, then Connect and sign in.

Team and Enterprise: an owner adds it first under Organization settings → Connectors → Add, hovers Custom, chooses Web, and pastes the URL. Members then connect it from their own Settings → Connectors.

https://adako.ai/mcp

The number of custom connectors you can add depends on your Claude plan.

Claude Desktop

Same Connectors settings as Claude.ai — add the custom connector with the URL above, then sign in when prompted. Desktop shares your Claude account, so a connector added on the web is available here too. Full guide: clients/claude-desktop.md.

Claude Code

claude mcp add --transport http adako https://adako.ai/mcp

Then run /mcp inside a session and pick adako to sign in through the browser.

Add it once for every project with --scope user:

claude mcp add --transport http adako https://adako.ai/mcp --scope user

API-key variant, no browser step:

claude mcp add --transport http adako https://adako.ai/mcp \
  --header "Authorization: Bearer ak_live_your_key"

There is also a plugin that installs the server, the skills, a setup command and an operator agent in one step:

/plugin marketplace add ruslan2027/adako-mcp
/plugin install adako@adako
/adako-setup

See clients/claude-code.md and plugin/.

ChatGPT (developer mode)

1. Turn on Developer mode under Settings → Security & login, then open ChatGPT Plugins.
2. Click +. Enter Adako as the connection name and https://adako.ai/mcp as the MCP server URL. Add a short description, choose the public endpoint connection option, and click Create.
3. Complete authorization if prompted.
4. Start a new conversation, open the tools menu, and select Adako.

Full guide: clients/chatgpt.md.

Cursor

One command, idempotent, and it leaves your other servers alone:

node install/cursor.mjs            # the home-folder .cursor/mcp.json
node install/cursor.mjs --project  # .cursor/mcp.json in this repository

Or write the file yourself:

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

Full file: clients/cursor/mcp.json. Guide: clients/cursor.md.

This repository is also a Cursor plugin, for the Cursor Marketplace: .cursor-plugin/plugin.json adds the server and the skills in skills/.

Codex CLI

codex mcp add adako --url https://adako.ai/mcp
codex mcp login adako

Or run the installer, which rewrites one table in the Codex config (C:\Users\<you>\.codex\config.toml on Windows or /Users/<you>/.codex/config.toml on macOS) and leaves the rest of the file untouched:

node install/codex.mjs            # API key from $ADAKO_API_KEY
node install/codex.mjs --oauth    # sign in with codex mcp login instead

Full file: clients/codex/config.toml. Guide: clients/codex.md.

Windsurf

node install/windsurf.mjs

Or add the block by hand to C:\Users\<you>\.codeium\windsurf\mcp_config.json on Windows or /Users/<you>/.codeium/windsurf/mcp_config.json on macOS:

{
  "mcpServers": {
    "adako": {
      "serverUrl": "https://adako.ai/mcp"
    }
  }
}

Full file: clients/windsurf/mcp_config.json. Guide: clients/windsurf.md.

Gemini (custom app)

On a computer, open gemini.google.com/apps, or Settings → Connected apps, then scroll to Custom apps at the bottom, choose Add a custom app, and paste the URL:

https://adako.ai/mcp

Click Next and sign in. Leave the credential fields under Advanced features empty: Adako registers itself. In a chat, type @ and pick Adako.

Guide: clients/gemini.md. Custom apps are still rolling out, with conditions Google lists: a personal Google Account, English, the United States, 18 or over.

Antigravity (app, IDE and the agy CLI)

All three read one file, so Adako is added once. In the desktop app: Settings → Customizations, then Open MCP Config. In the IDE: the agent side panel, … → MCP Servers. In the CLI: /mcp, or the file itself — C:\Users\<you>\.gemini\config\mcp_config.json on Windows or /Users/<you>/.gemini/config/mcp_config.json on macOS (Linux: /home/<you>/.gemini/config/mcp_config.json). Use .agents/mcp_config.json for one project. After saving, press refresh beside Installed MCP Servers and click Adako to sign in.

{
  "mcpServers": {
    "adako": {
      "serverUrl": "https://adako.ai/mcp"
    }
  }
}

Antigravity signs in through the browser on the first call. Guide: clients/antigravity.md.

Gemini CLI stopped serving free and Pro accounts on 18 June 2026 and now runs only on paid Gemini and Gemini Enterprise API keys. If you are still on it, the older spelling works: httpUrl in .gemini/settings.json in your home folder, with the extension manifests gemini-extension.json and clients/gemini-cli/gemini-extension.json. Separately, agy plugin import gemini carries an existing setup into Antigravity.

VS Code (Copilot)

.vscode/mcp.json in a workspace, or your user mcp.json:

{
  "servers": {
    "adako": {
      "type": "http",
      "url": "https://adako.ai/mcp"
    }
  }
}

VS Code handles the OAuth flow. Full file with the API-key variant: clients/vscode/mcp.json. Guide: clients/vscode.md.

Perplexity

Add Adako as a custom connector: on the web at perplexity.ai/computer/connectors, or Settings → Connectors in the desktop app, then the URL above with OAuth. Custom connectors come with Pro, Max and Enterprise. Guide: clients/perplexity.md.

Add the skills

skills/ holds eleven Agent Skills. adako is the operating procedure: the safety contract, the tool-call contract, the tool map and the workflows. The others are focused: one per platform, one per job.

• Claude Code — install the plugin, or copy skills/ into the .claude/skills/ folder in your home directory (C:\Users\<you>\.claude\skills\ on Windows or /Users/<you>/.claude/skills/ on macOS).
• claude.ai — upload SKILL.md (the same file as skills/adako/SKILL.md) as a skill in settings.
• Anything else — paste the contents into your system prompt or project instructions.

The server also ships short instructions on connect, so the skills are optional. They make long sessions noticeably steadier.

REST API and the command line

The same tools run over HTTP with an API key.

curl -X POST https://adako.ai/api/v1/tools/google_get_campaign_performance/execute \
  -H "Authorization: Bearer ak_live_your_key" \
  -H "Idempotency-Key: $(uuidgen)" \
  -H "Content-Type: application/json" \
  -d '{"arguments": {"date_range": "last_30_days"}}'

GET /api/v1/tools lists every callable tool. GET /api/v1/openapi.json is the machine-readable contract, which n8n, Zapier and Make can import. Full reference: adako.ai/docs/api.

The adako command line wraps the same endpoints for scripts and coding agents:

npx @adako/cli login
npx @adako/cli google list-campaigns --customer-id 1234567890
npx @adako/cli meta get-campaign-performance --date-range last_30_days --raw

Its source is in cli/: plain ESM, no dependencies, Node 22 or newer. The published package is @adako/cli on npm.

Connect your accounts

1. Sign in at adako.ai.
2. Open Accounts and connect a platform. Google Ads and LinkedIn Ads sign in through the platform; Meta Ads takes an access token from a system user in your own Meta app; ChatGPT Ads takes an advertiser API key. You never hand a password or a token to your AI client.
3. Turn on the accounts Adako may work on, and pick the primary for each platform.
4. Back in your assistant, say "start here".

First prompts

Read-only, so nothing can move:

• "Start here."
• "Which ad accounts am I connected to?"
• "How did my Google campaigns do in the last 30 days compared with the 30 before?"
• "Find search terms that spent money and converted nothing last month."
• "Which Meta ad sets are spending without results this week?"
• "Break my best Meta ad set down by placement."
• "Which of my TikTok ads are showing creative fatigue?"
• "What did LinkedIn cost per lead last month, and which job titles did it reach?"
• "Run that wasted-spend check across all my active accounts."

Then, when you want a change:

• "Suggest ten negative keywords from that list and tell me what each one blocks."
• "Pause the campaign that spent the most with no conversions."
• "Raise the daily budget on Brand Search to 40."
• "Draft a paused Search campaign for {product} targeting {city}, 25 a day."
• "Alert me if CPA goes above 80 two days running, and send me a brief every Monday."

Every one of those ends with a proposal you approve or reject.

Pricing

Free includes 1 ad account and 30 tasks a month. Pro ($49/month) covers 5 ad accounts and 600 tasks a month, Agency ($149/month) 20 accounts and 5,000 tasks, and Enterprise is priced per contract. An account used in a billing period keeps its place until the period resets. A task is one billed change to an ad account; reads and connection, usage, proposal, discovery, diagnostic and resolver tools are free. Monitors, scheduled briefs and generated reports need Pro or above. Details at adako.ai/pricing.

Docs

adako.ai/docs — quickstart, one page per client, one page per platform, the tool surface, workflows, monitors, the full tool reference and the REST API.

Support

support@adako.ai · security reports: see SECURITY.md · changes: see CHANGELOG.md

Licence

The skills and the client configurations in this repository are MIT — see LICENSE. The hosted Adako service is proprietary.