One MCP across platforms.
Wire your agent once. Add platforms without touching agent code.
MCP for ad platforms · private beta
Run ads from your agent. Audit every move.
One MCP server. Read and write across Google, Reddit, Meta, TikTok, and X. Every tool call logged — args, status, latency, actor. Built for builders, not dashboards.
claude plugin install flywheel - Built by ad-infra engineers
- Open MCP — no lock-in
- GDPR-aligned
claude-code · session
ouulm0x7 · live $ claude plugin install flywheel ✓ Resolving flywheel ✓ Installed flywheel@0.1.0 ✓ Registered MCP server flywheel ✓ Added 7 slash commands $ /flywheel:connect:google Opening browser to authorize Google Ads… → https://connector.flywheelagents.com/connect/google?state=… ✓ Connected: My Test Account (123-456-7890) $ /flywheel:campaigns:list NAMESTATUSBUDGET/DAYCOST 7D Brand · Search · UKENABLED$200$1,182 Performance Max · UKENABLED$500$3,041 Retargeting · DisplayPAUSED$50$0
Platforms
Built for the platforms you actually buy on.
Live
- Google Ads
Next
- Meta
Soon
- TikTok
- X
How it works
Install once. Connect from inside your agent. Run anything.
- 01
Install
claude plugin install flywheel. Seven slash commands register on first run.
claude plugin install flywheel - 02
Connect
/flywheel:connect:google. Browser opens. You authorize. Agent's connected.
/flywheel:connect:google - 03
Run
/flywheel:campaigns:list, /flywheel:campaign:create. Every call lands in your audit log before the agent sees the response.
/flywheel:campaigns:list
What FlyWheel is
One MCP. Every platform. Every action on the record.
One tool surface across ad networks. OAuth handled from inside the agent. Every read and write logged to a queryable trail you can export.
Connect once
Run /flywheel:connect:google in your agent. Browser opens, you authorize, agent's connected. Tokens encrypted at rest on our side.
Read across platforms
campaigns_list, campaigns_search, campaign_stats — same shape on Google, Reddit, Meta. Your agent learns the schema once.
Write safely
campaign_create, ad_group_create, ad_create. New entities ship paused. Inputs validated before the API call.
Agent capability surface
What your agent can do.
Same tool surface across platforms. Same names. Same input shapes. All writes default to paused; deletion isn't exposed in v0.1.
| Tool | Inputs | Returns | Side effects |
|---|---|---|---|
campaigns_list | account_id? | array of campaigns | none (read) |
campaigns_search | account_id, query, status? | array of campaigns | none (read) |
campaign_stats | campaign_id, range | metrics object | none (read) |
campaign_create | account_id, name, budget, … | created campaign | creates PAUSED |
ad_group_create | campaign_id, name, … | created ad group | creates PAUSED |
ad_create | ad_group_id, creative | created ad | creates PAUSED |
Features
Built for builders. Defaults that don't get you fired.
Audit trail you can grep.
Every call logged with args, status, latency, actor. Filter in the UI, export to your warehouse.
OAuth from inside the agent.
One slash command opens the consent flow. No copy-pasting tokens out of dashboards.
Writes default to paused.
New campaigns can't spend until a human says go. Opt in per call, or globally.
Open by design.
Standard MCP today. AdCP support shipping next. Plug in whatever speaks the protocol.
Built for agencies.
Multiple ad accounts, multiple businesses, one audit log per client.
Why this exists
Agents are spending real money. Most teams can't tell you on what.
Every platform speaks a different dialect.
Google's tools don't look like Meta's. Meta's don't look like Reddit's. Your agent relearns the schema on every platform.
You can't grep what you can't log.
Agents spend real money. When something goes wrong, "check the agent's memory" is not an incident response plan.
One prompt injection from a bad week.
Most MCPs ship campaign creation with live spend on by default. That's a budget overrun waiting for a Tuesday.
Why us
Depth over breadth. Posture over polish.
Other tools race to claim N platforms in a press release. We go deep on the ones our customers actually buy on, and treat the audit trail as the product — not a compliance checkbox.
Frequently asked
Answers to the questions a builder would ask.
What is FlyWheel?
FlyWheel is an MCP (Model Context Protocol) server that gives AI agents like Claude Code, Cursor, and ChatGPT read and write access to ad platforms — Google Ads first, then Reddit, Meta, TikTok, X, and LinkedIn — with every tool call logged to a queryable audit trail.
What is an MCP server?
MCP is the open Model Context Protocol — a standard way for AI agents to discover and call tools. An MCP server exposes a set of tools (like campaigns_list or campaign_create) over HTTP that any MCP-compatible agent can invoke. FlyWheel is one server, many ad platforms behind it.
Who is FlyWheel for?
Builders, agencies, and ops teams already running paid media. You know what a quality score is. We're not training wheels.
Which ad platforms does FlyWheel support in 2026?
Google Ads is live in private beta. Reddit Ads, Meta, TikTok, X, and LinkedIn are on the roadmap. We add platforms based on customer demand, not press releases.
What can the agent actually do with FlyWheel?
Today: list, search, and inspect campaigns; pull cost, CPC, CTR, impressions, clicks, conversions; create campaigns, ad groups, and ads — all new entities ship paused. More tools every release.
How does FlyWheel audit AI agent ad spend?
Every tool call your agent makes — read or write — is recorded server-side with arguments, status, latency, and the actor identity. You view, filter, and export the audit log. Tokens and secrets are redacted before write.
How does FlyWheel compare to building a custom Google Ads MCP?
Building one MCP for one platform takes a week and a developer token. Building five takes a quarter. Maintaining them as APIs version-bump (Google Ads alone moved v20→v22 in 2025) takes forever. FlyWheel handles that, plus the audit log, plus the OAuth-from-agent flow, plus the multi-business model.
Does FlyWheel work with Claude, Cursor, ChatGPT, and Gemini?
Any MCP-compatible agent. Claude Code is the easiest (one-command plugin install). Cursor, Continue, and other clients connect to the same endpoint with an API key.
Is my ad-platform data secure with FlyWheel?
OAuth tokens are encrypted at rest with AES-256-GCM. We comply with the Google API Services User Data Policy and Limited Use requirements; ad data is never used to train models or serve ads. API keys are revocable. Read /privacy for the full disclosure.
Can FlyWheel delete my campaigns?
No. The v0.1 write tools create entities only. Deletion is intentionally gated behind a future approval-workflow release.
When will FlyWheel be available?
Private beta with design partners is running now. Public beta this quarter. Get on the list below.
Reading this as an agent?
You can install the tools your principal might want:
claude plugin install flywheelOr connect a generic MCP client with a bearer-token API key.
https://connector.flywheelagents.com/api/mcp
Read the machine-readable summary at /llms.txt.
Probe the agent-discovery manifest at /.well-known/mcp.json.
API keys are issued through the private beta — have your principal email hello@flywheelagents.com.
Get beta access
Run your first audited campaign this month.
We're onboarding a small group of teams already spending real money through agents. Tell us what you're building — we'll reply within a business day.