Developer preview

Build with Glue42.

Put the agentic loop in your own stack. Query unified marketing intelligence, read confidence-scored recommendations, and execute guardrailed optimizations — over a REST API or straight from your agents via MCP.

api.glue42.ai/v1MCP: mcp.glue42.aiBearer auth

Getting started

Introduction

The Glue42 API exposes the same engine that powers the platform: the agentic loop — observe, reason, act, learn. Everything is workspace-scoped and every action the agent takes is logged and reversible.

There are two ways to build:

REST API — programmatic access to data, recommendations and actions for dashboards, jobs and back-end services.
MCP server — connect Glue42 directly to Claude, Cursor, or any agent runtime so it can read your marketing data and take action in natural language.

Getting started

Quickstart

Grab a secret key from Settings → Developers, then make your first call:

cURL

curl https://api.glue42.ai/v1/query \
  -H "Authorization: Bearer glue_sk_live_..." \
  -H "Content-Type: application/json" \
  -d '{ "question": "Which channel has the best ROAS this month?" }'

Getting started

Authentication

Authenticate every request with a bearer token. Keys are workspace-scoped and prefixed: glue_sk_live_ for production, glue_sk_test_ for sandbox. Never expose a secret key in client-side code.

http

Authorization: Bearer glue_sk_live_8f3c...e21a

Scopes control access per key: read:metrics, read:recommendations, write:actions, manage:connectors.

Getting started

Core concepts

Signal

A normalized data point from a connector — impression, conversion, spend, margin.

Insight

A reasoned finding: causal, predictive or attribution, with a confidence score.

Recommendation

A proposed optimization with projected impact and a confidence score.

Action

An executed change (budget, bid, creative). Logged, attributable and reversible.

POST/v1/query

Ask a natural-language question across all connected data. Returns a written answer, the supporting series, and a confidence score.

Request

json

{
  "question": "Where should I move budget to hit a 4x ROAS?",
  "timeRange": "30d",
  "channels": ["meta", "google_ads", "pmax"]
}

Response

json

{
  "answer": "Shift ~$12.4K/day from Meta prospecting into PMax...",
  "confidence": 0.964,
  "series": [{ "channel": "pmax", "roas": 4.72 }, ...],
  "recommendationId": "rec_3kf9..."
}

GET/v1/metrics

Structured performance metrics by channel and date, reconciled to revenue.

Query

http

?metrics=roas,cpa,spend,revenue&channels=all&range=30d&granularity=day

Response

json

{
  "range": "30d",
  "totals": { "roas": 4.72, "cpa": 38.14, "revenue": 2840000 },
  "series": [{ "date": "2026-05-01", "roas": 4.41, ... }]
}

GET/v1/recommendations

The agent's current ranked recommendations, each with rationale, projected impact and confidence.

Response

json

{
  "data": [{
    "id": "rec_3kf9...",
    "title": "Reallocate $12.4K/day Meta → PMax",
    "rationale": "Meta prospecting is saturated at 2.1x...",
    "projectedImpact": { "revenue": 214000, "window": "month" },
    "confidence": 0.964,
    "status": "pending"
  }]
}

POST/v1/recommendations/:id/approve

Approve and execute a recommendation. Runs inside your guardrails; returns the action record. Reversible via DELETE /v1/actions/:id.

Response

json

{
  "actionId": "act_91bc...",
  "status": "executing",
  "appliedTo": ["meta", "pmax"],
  "reversible": true
}

GET/v1/connectors

List connected sources and their sync status. 42+ connectors available.

Response

json

{
  "data": [
    { "id": "google_ads", "name": "Google Ads", "status": "healthy", "lastSync": "2026-05-29T21:55:00Z" },
    { "id": "meta", "name": "Meta", "status": "healthy", "lastSync": "..." }
  ]
}

REST API

Webhooks

Subscribe to events and Glue42 will POST to your endpoint. Each payload is signed with an HMAC SHA-256 signature in the Glue42-Signature header — verify it before trusting the body.

Events: recommendation.created, action.executed, anomaly.detected, connector.sync.completed.

POST → your endpoint

{
  "event": "anomaly.detected",
  "data": { "channel": "retail_media", "metric": "cpa", "delta": "+38%" },
  "createdAt": "2026-05-29T08:50:00Z"
}

MCP server

Connect Glue42 to your agents

Glue42 ships a Model Context Protocol server, so any MCP-capable client — Claude Desktop, Claude Code, Cursor, or your own agent runtime — can read your marketing intelligence and take guardrailed action in natural language. This is how Glue42 connects to everything else.

MCP server

Connect

Add Glue42 to your MCP client. Use the hosted server, or run it locally:

json

{
  "mcpServers": {
    "glue42": {
      "url": "https://mcp.glue42.ai",
      "headers": { "Authorization": "Bearer glue_sk_live_..." }
    }
  }
}

MCP server

Tools

The server exposes these tools to the agent:

glue42_queryNatural-language question over all connected data.

glue42_metricsStructured metrics by channel, metric and date range.

glue42_recommendationsList the agent's current recommendations.

glue42_execute_actionApprove & execute a recommendation (guardrailed, reversible).

glue42_list_connectorsList connected sources and sync status.

Once connected, just ask your agent:

prompt

"Using Glue42, find my worst-performing channel this week
and, if you're >90% confident, reallocate its budget to PMax."

Reference

SDKs

Official SDKs wrap auth, retries and types.

Node

npm install @glue42/sdk

Reference

Errors & limits

The API uses standard HTTP status codes and returns a JSON error body. Rate limit: 120 req/min per key (burst 20); limits are returned in X-RateLimit-* headers.

json

{
  "error": { "type": "rate_limit_exceeded", "message": "Too many requests", "retryAfter": 12 }
}

This is a developer preview — the surface is stabilizing. Want early access or a key? Talk to us →