Skip to main content

AI Overview

Agent Flow schema and MCP server are preview features

Agent flow schema and hosted MCP servers are preview features that require Prismatic Support to enable the feature for you. Please contact Prismatic Support if you'd like to opt-in to the MCP Preview.

Agent Flows, AI Components, and MCP acceleration

Build AI-powered integrations

Transform your product's integrations with built-in intelligence. Prismatic lets you drop in AI connectors, orchestrate agent flows across systems, and give your AI agents live context via the Model Context Protocol (MCP). You don't need to rebuild your integration layer to launch smart features quickly.

Ship AI Features Faster

Connect to OpenAI, ChatGPT, and other AI services through ready-to-use components. Focus on building the integration logic that matters to your customers while we handle the AI service plumbing

Request AI preview

What you can build

Explore the AI capabilities below. Each one includes a quick example to show how it works, with links to more detailed documentation if you want to go deeper.

Use built-in AI components (Preview)

Ready-to-use connectors for services like OpenAI and Anthropic let you classify text, enrich data, and generate content. You can configure them in the UI or call them from code.

export const processTranscript = flow({
name: "ProcessTranscript",
description: "Process incoming transcripts and extracts key insights.",
onExecution: async (context, params) => {
const conversation = await context.components.googleGemini.sendMessage({
connection: context.configVars["Google Gemini Connection"],
model: "gemini-pro",
prompt: `Summarize this message and extract key entities: ${params.onTrigger.results.body.data}`,
});
return { data: conversation };
},
});

Explore AI Components →

Bring your own AI

Use your preferred agent SDKs and frameworks to integrate with any AI service or model. Build code-native flows with tools like LangChain, CrewAI, AutoGen, or connect to proprietary and on-prem models.

// Example: OpenAI agents SDK in a Prismatic flow
import { flow } from "@prismatic-io/spectral";
import { Agent, run, setDefaultOpenAIKey } from "@openai/agents";

export default flow({
name: "AI Agent Flow",
onExecution: async ({ configVars }, params) => {
// Set OpenAI API key from config
const apiKey = configVars.openAiApiKey.fields.apiKey;
setDefaultOpenAIKey(apiKey);

// Create a simple agent
const agent = new Agent({
name: "Assistant",
instructions: "You help users with their requests",
});

// Run the agent with user input
const result = await run(agent, [
{ role: "user", content: params.message },
]);

return { data: result.finalOutput };
},
});

Code Native Guide →

Expose integrations as AI tools with Agent Flows

Agent flows let you build workflows that can make decisions, call APIs, and interact with external systems autonomously. They're ideal for routing data, validating inputs, or triggering the right actions at the right time.

import { type CheckCalendarEventsInput } from "../schemas/flows";
import { createGraphClient, extractAccessToken } from "../services/graphClient";

export const checkCalendarEvents = flow({
name: "Check Calendar Events",
description:
"Check upcoming calendar events for a specified number of days ahead",
schemas: {
invoke: {
properties: {
daysAhead: {
type: "number",
description: "Number of days ahead to check",
},
includeAllDay: {
type: "boolean",
description: "Include all-day events",
},
},
required: ["daysAhead", "includeAllDay"],
},
},
onExecution: async (context, params) => {/* workflow logic */};
});

Agent flow schema →

Combine AI with human oversight

For critical workflows, combine automation with human approval to ensure accuracy and maintain trust. Approval tasks can be added to agent flows to pause and wait for review before proceeding.

Incident Monitoring

Reference Examples:

  • Slack Chatbot Agent – Build interactive AI agents that request human approval via Slack before taking actions. See the complete example on GitHub.
  • Human Approval Tools – Use the OpenAI Agent component to create custom human-in-the-loop tools for your workflows. Learn how to create human approval tools.

Accelerate with MCP (Preview)

The Model Context Protocol (MCP) connects AI agents to your integration platform. Choose the right MCP server based on your use case:

Two ways to use MCP:

  • Building with AI – Use the Prism MCP Server in your IDE to accelerate integration development. Connect your AI copilot (Cursor, Windsurf, etc.) to generate code, publish integrations, and reference components directly from your development environment. Get started with Prism MCP →

  • Running AI in Production – Use the Prismatic MCP Server to connect your production AI agents to customer integrations. This exposes your agent flows as tools that AI agents can invoke to interact with external systems and data.

    {
    "mcpServers": {
    "prismatic": {
    "command": "npx",
    "args": ["mcp-remote", "https://mcp.prismatic.io/mcp"]
    }
    }
    }

MCP documentation →

Common AI use cases

Explore practical patterns for implementing AI capabilities in your integrations. Each example includes working code and detailed implementation guides.

Data Enrichment

Add context and insights to existing data using AI analysis. Research entities, analyze sentiment, score records based on custom criteria, or augment data with external information.

See lead enrichment example →

View all implementation details in Common Use Cases →

How to get started

Follow these steps to build your first AI-enabled integration:

  1. Start with a template – Choose an AI-enabled template that includes common patterns and best practices. Browse templates →

  2. Add AI components – Insert a built-in component or create a custom one to add intelligence to your flows. Add AI components →

  3. Experiment with agent flows – Build autonomous workflows that can decide and act. Agent flow guide →

  4. Enable MCP – Use MCP servers to provide your AI agents with context and accelerate development. MCP setup →

Request AI preview →

Best practices and considerations

To build reliable, secure AI integrations, keep these guidelines in mind:

  • Pick the right approach – Start with built-in components for common tasks; use custom connectors for specialized models; use agent flows for multi-step processes.
  • Provide context and handle errors – Use structured output for deterministic AI responses, implement fallback logic, and use timeouts where appropriate.
  • Monitor and control usage – Track API usage and costs; set rate limits and implement throttling.
  • Secure your data – Never expose API keys in client-side code; use environment variables for secrets; sanitize inputs and follow AI service security best practices.

If you need more guidance, consult our documentation on custom connector development, integration patterns, and security configuration.

Frequently Asked Questions

Prismatic MCP Server runs in production and gives AI agents access to your deployed integrations and data. Prism MCP Server runs in your IDE (like Kiro) and helps with development by providing AI-powered code generation and integration building assistance.

Learn more about MCP →

Your data is processed securely with encryption in transit and at rest. AI services only receive the specific data you configure to send, and you maintain full control over what information is shared with AI models.

Security documentation →

Yes! You can create custom components to connect to any AI service or model. The Spectral SDK provides the tools to build connectors for proprietary models, on-premises AI services, or specialized AI APIs.

Custom connector guide →

AI features including agent flows and MCP servers are currently in preview. Contact our team to request early access and get started with AI-powered integrations.

Request preview access →

Yes, we provide comprehensive examples for AI components, agent flows, and MCP server implementations. You'll find code samples in our documentation and GitHub repositories.

Browse examples →

Resources

AI Component Library

Browse built-in AI components and their capabilities

Learn more →