Agent SmithOverview

Agent Smith

SignalSmith’s Agent Smith brings AI-powered capabilities to your CDP operations, enabling faster audience building, conversational data exploration, and programmatic tool integration — all with built-in safety guardrails.

Capabilities

FeatureWhat It DoesWho It’s For
NL Audience BuilderDescribe an audience in plain English and get filter conditions generated automaticallyMarketing analysts, growth operators
AI AgentConversational interface that performs multi-step CDP operations on your behalfAll users — data engineers, analysts, operators
MCP ServerModel Context Protocol server exposing SignalSmith as AI-accessible toolsDevelopers, AI engineers, automation builders
GuardrailsSafety boundaries that control what AI operations can do without human approvalPlatform admins, security teams

How It Works

The Agent Smith is built on three layers:

1. MCP Server (Tool Layer)

The MCP Server exposes SignalSmith’s capabilities as structured tools that any MCP-compatible AI client can invoke. It provides read tools (list models, get audiences, introspect sources) and write tools (create audiences, trigger syncs, configure destinations). Every tool call goes through the same authentication and authorization as the REST API.

2. Agent Service (Intelligence Layer)

The Agent Service runs AI-powered agent loops that can chain multiple tools together to accomplish complex tasks. When you ask “Create an audience of users who spent over $500 in the last 90 days,” the agent:

  1. Introspects your data sources to find relevant tables and columns
  2. Identifies or creates the appropriate traits
  3. Generates filter conditions
  4. Presents the result for your review

The agent delivers real-time progress updates as it works.

3. Guardrails (Safety Layer)

Guardrails enforce policies on AI operations. You define rules like “require human approval before deleting an audience” or “block AI from modifying production syncs.” Guardrails apply to all AI-initiated actions regardless of which client (chat UI, MCP client, API) triggered them.

Getting Started

Using the AI Chat (Fastest)

  1. Click the AI button in the SignalSmith UI sidebar
  2. Type a natural language request (e.g., “Show me my largest audiences”)
  3. The agent responds with results and can take follow-up actions

Using the NL Audience Builder

  1. Navigate to Segment > Audiences
  2. Click Create with AI
  3. Describe your audience in plain English
  4. Review the generated filter conditions
  5. Adjust if needed and save

Connecting an MCP Client

For developers who want to use SignalSmith tools in their own AI workflows:

  1. Generate an API key in Settings > API Keys
  2. Configure your MCP client (Claude Desktop, Cursor, etc.) to connect to your SignalSmith MCP Server
  3. Start using SignalSmith tools in your AI conversations

See the MCP Server documentation for detailed setup instructions.