Build an AI agent

An AI agent stack combines a model that can use tools, sources of context, and APIs that perform actions. Start with a model provider, add retrieval or web search when your task needs fresh information, and connect only the business tools your agent needs.

Choose your building blocks

AI models

Anthropic↗

Build assistants that reason, write code, and use tools with the Claude API.

Language modelsTool use

Documentation ↗Source reviewed

Search & data

Firecrawl↗

Turn websites into structured data and readable content for AI applications.

Web scrapingExtraction

Documentation ↗Source reviewed

Search & data

Exa↗

Search the web and retrieve page content for AI research and agent workflows.

Web searchResearch

Documentation ↗Source reviewed

Developer tools

Supabase↗

Connect AI tools to your Supabase projects to query data and manage development workflows.

DatabaseMCP

Documentation ↗Source reviewed

Business APIs

Stripe↗

Give AI assistants access to Stripe tools for payments and business operations through MCP.

PaymentsMCP

Documentation ↗Source reviewed

Business APIs

Resend↗

Send transactional emails from your application with an API built for developers.

EmailNotifications

Documentation ↗Source reviewed

AI models

Cohere↗

Build enterprise AI with language models, embeddings, and reranking for retrieval workflows.

EmbeddingsReranking

Documentation ↗Source reviewed

Search & data

Pinecone↗

Store and search vector data to retrieve relevant context for AI applications.

Vector databaseRetrieval

Documentation ↗Source reviewed

Search & data

Tavily↗

Search the web and extract content for research agents and AI applications.

Web searchResearch

Documentation ↗Source reviewed

AI models

Groq↗

Run supported language and audio models through Groq’s inference API.

InferenceAudio

Documentation ↗Source reviewed

Developer tools

Notion↗

Connect AI assistants to workspace content using Notion’s hosted MCP server.

KnowledgeMCP

Documentation ↗Source reviewed

What to evaluate

• Evaluate model quality on tasks from your own application.
• Check tool permissions, authentication, and approval requirements before enabling writes.
• Measure the total cost and latency of a complete task, including tool calls.

Provider documentation

Read the original sources for current capabilities, setup requirements, and usage terms. These selections are starting points, not a performance ranking.

Keep exploring