Toolhouse

Toolhouse is a platform for building, deploying, and running AI agents without the usual boilerplate. In its Agent Studio, you describe what you want an agent to do in plain language, for example scraping a URL and summarizing it, and Toolhouse assembles the agent, wires up the tools it needs, and turns it into a scalable, production ready API you can deploy in one click. It is pitched at both developers, who can drop its SDK into an existing stack, and less technical builders, who can ship working agents without writing code. The platform is used by companies including Cloudflare, NVIDIA, Groq, and Snowflake.

The core value is that Toolhouse handles the backend an agent needs out of the box: tool integrations, retrieval augmented generation with automatic setup, memory, caching, logging, debugging, and deployment. It ships a library of prebuilt tools such as web search through Google, Tavily, and Exa, scraping through Firecrawl, databases like MongoDB, Pinecone, and Supabase, and email through SendGrid. Its standout feature is MCP support. Agents connect to thousands of remote Model Context Protocol servers with a single link, and MCP Discovery reads your prompt, detects which MCP servers the agent needs, and configures them automatically, removing the usual setup friction. Bundles let you control exactly which tools an agent can use.

For developers, the Toolhouse SDK provides tool definitions directly to LLM calls and works with OpenAI, Vercel AI, and LlamaIndex, with Python and TypeScript libraries and a CLI. Each agent gets a sandbox for testing and debugging, then deploys to a production grade cloud with a unique API endpoint that can be called from applications or services like Zapier. On security, data is encrypted in transit and at rest, is never used to train models, and stays scoped to your workspace, and every agent action is logged for audit. Enterprise plans add on premises deployment, private data hosting, and compliance assistance.

Pricing is deliberately bootstrapper friendly. A free Sandbox tier costs nothing and includes 50 agent runs a month with unlimited MCP calls, suited to getting started and public projects. The Pro plan is $10 a month and adds 1,000 agent runs a month, private agents, built in RAG, and the ability to switch between models. Enterprise is custom and adds dedicated support, on premises deployment, private data hosting, and compliance help. An agent run is counted each time you call an agent to perform an operation and costs one credit, with default model LLM usage and unlimited MCP executions included in each run.