SFMC MCP Server

One-Line Summary: A zero-dependency Model Context Protocol server that gives AI agents direct, authenticated access to Salesforce Marketing Cloud. turning natural language into live SFMC queries.

--

Problem Statement

Salesforce Marketing Cloud (SFMC) is notoriously difficult to navigate. Querying assets, checking journey statuses, or inspecting automations requires clicking through a slow web UI or writing custom API integration code from scratch every time. For a team using AI coding agents, there was no way to say "show me all active journeys" and get a live answer. The AI agents had no connection to SFMC. they could only guess based on documentation, not query the actual tenant.

Solution

Built a custom MCP (Model Context Protocol) server that bridges AI agents to the SFMC REST API. The server implements the MCP stdio transport protocol, authenticates via SFMC's server-to-server OAuth 2.0 flow, and exposes a set of tools that any connected AI agent can call directly from natural language prompts. The entire server is a single Python file with zero external dependencies. only the Python standard library is used.

The server is designed to be team-sharable: the .mcp.json project config file can be committed to any repo, and each teammate only needs to create their own .env with SFMC credentials.

Tech Stack

Key Features

• Zero External Dependencies: The entire server runs on Python's standard library (urllib, json, sys, dataclasses). No pip install step, no virtual environment, no dependency conflicts. Copy the file and run it.

• Six Capabilities Exposed to the AI Agent:

• Inspect non-secret configuration and safety flags before any call

• Fetch and validate an SFMC access token on demand

• Execute arbitrary REST API calls against the SFMC tenant

• Browse Content Builder assets with pagination

• List all journeys with pagination

• List all automations with pagination

• Read-First Safety Model: GET requests work by default. All mutating HTTP methods (POST, PATCH, PUT, DELETE) are blocked unless the SFMC_ENABLE_MUTATIONS=true environment variable is explicitly set. This prevents AI agents from accidentally modifying production marketing assets.

• Token Caching with Auto-Refresh: Access tokens are cached in memory and automatically refreshed 60 seconds before expiry. No redundant auth calls.

• Full MCP Protocol Compliance: Implements initialize, tools/list, tools/call, and notifications/initialized handlers with proper JSON-RPC 2.0 error codes.

• Team-Sharable via .mcp.json: The project config file registers the server at the project level. Any teammate who opens the project gets prompted to approve the MCP server. no manual CLI setup.

• Validated Against Live Tenant: Successfully tested against the client's production SFMC instance. token acquisition, endpoint enumeration, asset listing, journey listing, and automation listing all confirmed working.

Impact / Metrics

• Replaced a multi-click web UI workflow (login, navigate, drill down, scroll, export) with a single natural-language query. Asset, journey, and automation lookups that took 3-5 minutes manually now run in under 10 seconds, with the result delivered as structured data the AI agent can act on directly.
• Enabled AI-assisted marketing cloud management. agents can now inspect journeys, assets, and automations without human navigation
• Zero onboarding friction for teammates: no dependencies to install, just add credentials
• Safety-first design prevents accidental mutations in production

Status

Production. Successfully tested against live SFMC tenant. Server is stable and in use for day-to-day SFMC queries via AI agent. Future planned tools include data extension queries, subscriber lookups, and send log inspection.