Model Context Protocol Server

What is the Model Context Protocol?

The Model Context Protocol (MCP) is an open standard that lets AI clients — Claude Desktop, ChatGPT Desktop, Cursor, Windsurf, OpenClaw, Cline, Zed, and Claude Code — connect to external data sources through a uniform tool-calling interface. Instead of every AI vendor inventing its own integrations, MCP defines a single contract: the server exposes tools, the client calls them, and the model sees the results inside its context window.

For an analytics platform, MCP is the missing piece. It means you can sit inside the AI client you already pay for — the one with your favorite reasoning model, the one with your saved prompts, the one your team has standardized on — and pull live revenue, cohort, MER, LTV, and campaign data from your Admaxxer workspace as if those numbers lived inside the chat. No copy-paste. No screenshots. No exporting CSVs to feed back into the model.

Why this matters for DTC analytics

The dominant DTC analytics tools (Triple Whale, Northbeam, Hyros) ship their own AI agents tied to their own model choice — usually a single fine-tune they fund and bill you for. You get one model, one vendor, one inference budget. If the model's reasoning isn't strong enough for the question you're asking, your only options are "wait for the vendor's next release" or "leave the platform and take the data with you."

Admaxxer is the inverse. The data lives here. The intelligence lives wherever you want — Anthropic's Claude Sonnet 4.6, OpenAI's GPT-5, Google's Gemini 2.5 Pro, xAI's Grok 4, DeepSeek R2, Mistral Large, or any model that ships an MCP-capable client. You pay the model vendor directly at their list price. We don't mark up inference. We don't lock you into a model choice. The MCP server is the layer that makes that possible.

The six read-only tools

The MCP server exposes six tools — all read-only, all workspace-scoped, all audit-logged. Destructive campaign actions (pause, scale, launch) are deliberately not exposed via MCP; those stay inside the Admaxxer app behind explicit confirmation. See the ad operator vs. ads CLI doc for the reasoning.

1. list_connections

Returns the ad platform connections in the workspace and their health. Useful as the agent's first call to orient itself — "which accounts can I see?"

{ "tool": "list_connections" }
// → [{ platform: "meta", account_id: "act_123", status: "active", last_sync: "..." },
//    { platform: "google", account_id: "987-654-3210", status: "active", last_sync: "..." }]

2. list_campaigns

Lists campaigns across the connected ad platforms, with name, status, daily budget, and lifetime spend. Filter by platform or by date range.

{ "tool": "list_campaigns", "args": { "platform": "meta", "status": "active" } }

3. get_campaign_insights

Returns the insights timeseries for a single campaign — spend, impressions, clicks, conversions, ROAS, CPA — bucketed by day. Honors the workspace's primary attribution model (default: 7d-click + 1d-view).

4. get_account_insights

Same shape as get_campaign_insights but at the account level — useful for "what's our blended Meta ROAS this week" without enumerating every campaign.

5. run_analytics_queries

The headline tool. Calls one of the 33+ Tinybird pipes documented in the metric glossary — revenue, cohort LTV, blended MER, CAPI match rate, MMM contribution, forecast, incrementality. The tool enforces an allowlist (PIPE_ALLOWLIST in server/lib/tinybird/) so the agent can only read pipes that are safe to expose; experimental pipes are blocked at the server boundary.

{ "tool": "run_analytics_queries", "args": { "pipe": "blended_mer", "window": "30d" } }

6. read_workspace_context

Returns the operator-defined context document — your brand voice, your forbidden creative angles, your target CAC, your seasonal moments. The agent reads this before answering substantive questions so its recommendations land in your business reality, not in a generic "here's what the data says" vacuum.

Token model and security posture

Every MCP token is workspace-scoped. There is no "personal" or "global" token — the token belongs to a workspace, and a workspace's owner can revoke it at any time from Settings → Integrations → MCP. Revocation is immediate (cache TTL is 30 seconds). Tokens are stored encrypted at rest using AES-256-GCM with the same ENCRYPTION_KEY the rest of the platform uses (see the BYOK encryption doc).

Every tool call is audit-logged: timestamp, tool name, arguments hash, response size, and originating IP are written to the workspace's audit log. The owner can review every MCP call the same way they review every UI action. There is no "stealth mode" for AI access.

We deliberately do not expose destructive tools via MCP. The Claude Agent inside the Admaxxer app exposes 5 destructive-gated campaign tools (update_campaign, create_campaign, pause_all_low_roas) behind explicit confirmed: true user confirmation. Those tools are not in the MCP surface because the trust model of MCP — token in a remote client we don't control — is fundamentally weaker than a logged-in session in our own UI. A leaked MCP token can read your data. It cannot pause your campaigns or launch new ones. That is a deliberate ceiling, not an oversight.

Supported AI clients

Setup guides for each client live under /documentation/connect-any-ai/. Eight clients are first-class supported — meaning we test the full handshake, the auth flow, and the tool surface against the latest stable build of each client before every Admaxxer release:

• Claude Desktop — setup guide
• ChatGPT Desktop — setup guide
• Cursor — setup guide
• Windsurf — setup guide
• OpenClaw — setup guide
• Cline — setup guide
• Zed — setup guide
• Claude Code — setup guide

Any other MCP-spec-compliant client should work; we just don't run it through CI. If your client is on the official MCP clients list and the handshake fails, file an issue — we usually have a fix shipped within a week.

Setup in 30 seconds

1. Sign in to Admaxxer and open Settings → Integrations → MCP.
2. Click Generate token. The token is shown once; copy it immediately.
3. In your AI client's MCP config, add a server entry pointing at https://admaxxer.com/mcp with the token in the Authorization: Bearer <token> header.
4. Restart the client. The six tools should appear in the tool picker.

Learn more

• Connect any AI — overview
• AI ad operator vs. ads CLI — design philosophy
• Metric glossary — every pipe the agent can call
• Bring Your Own Key AI — pair with the in-app chat agent

---

Methodology

As of: April 30, 2026.

All product claims on this page reflect the live behavior of the Admaxxer platform on the date above. Where a metric is cited (e.g., the 92% sustained CAPI match rate), the measurement window, sample size, and source are stated inline next to the figure. Pricing tiers and plan inclusions are documented on the pricing page; if you find a discrepancy between this page and the pricing page, the pricing page is canonical and we'd like to know — please email hello@admaxxer.com.

Comparisons to third-party products (Triple Whale, Northbeam, Hyros) reflect publicly documented behavior of those products as of the as-of date. We do not maintain inside knowledge of competitor roadmaps; if a competitor has shipped a feature that changes the comparison, please let us know and we will update this page.