Give your AI agent superpowers with Dreamtune

Your agent can write a research plan. To finish the job, it may need current sources, a page-reading tool, a model for a second pass, or a voice service to turn the result into audio.

Dreamtune gives an existing agent access to supported AI models, tools and GPU compute through one account, API key and prepaid balance. You choose what the agent may use. Dreamtune manages the upstream access and records the charges in your workspace.

Here is a practical way to connect it and complete a first task. The setup below follows the current Dreamtune MCP guide and quickstart, checked October 18, 2026.

What can you give your agent?

Start in the catalog. It shows supported models, tools and compute, with the available operations and rates. The useful combination depends on the task:

• Research: search for sources, read relevant pages and ask a model to synthesize the evidence.
• Content production: draft a script, then call a supported image or voice tool.
• Development: use a model for a bounded coding task and inspect the result in your existing development environment.
• Compute workloads: rent an available GPU for a job that needs it, then stop it when finished.

These are workflows you can assemble. Connecting Dreamtune does not automatically build or run them for you. Your agent or application controls the sequence, and each selected operation has its own input requirements and cost. Adding MCP tools also does not automatically replace the model powering your agent; choosing another model is a separate application integration.

Connect through hosted MCP

For an agent with a compatible MCP client, the hosted connection avoids installing a local server. Your client must support Streamable HTTP and custom request headers.

Open API keys and create a key in the workspace you want to use. Select the access that your task needs. Then add these settings in your agent’s MCP connection screen:

• Transport: Streamable HTTP.
• Server URL: https://mcp.dreamtunelabs.com/mcp.
• Header: X-Dreamtune-API-Key.
• Header value: your Dreamtune production API key, saved in the client’s protected credential settings.

The server URL belongs in MCP settings. It is not a sign-in webpage. Hosted MCP uses the supplied key rather than a browser OAuth flow. Client configuration differs, so use the client-specific connection guides for your agent.

If the client cannot supply the header, follow the local MCP guide. Local MCP launches through the Dreamtune CLI and needs its documented environment configuration.

Discover first, then run a small task

Begin with catalog discovery and schema inspection. Ask your connected agent:

Find the available web-search tools in Dreamtune. Inspect one tool’s required inputs and published pricing. Show me what you would call before running a paid search.

That establishes whether the agent can see tools and understand an operation. Next, give it a bounded job with an observable result:

Research one product using its official website. Use a search tool if needed, cite the pages you read, and write a five-bullet brief. Stop after the brief and report the services you used.

Choose a small task you can afford to run. Review the actual requests and charges in Logs and the activity in Runs. A convincing answer alone does not prove that the intended tool ran.

Use the SDK or CLI in your own application

If you are building the agent, the same platform also has a TypeScript SDK and terminal CLI. The quickstart specifies Node.js 22.20 or later. Install the package in a server application:

npm install @dreamtune/sdk

For terminal use, install the CLI:

npm install --global @dreamtune/sdk

Follow the quickstart to supply your API key, API origin and request timeout. Then this read-only command discovers search tools:

dreamtune tools list --q search

Use the identifiers returned by discovery and inspect the selected schema before execution. The SDK guide covers application calls; the tools guide covers discovery, inspection and running tools. Keep the key in protected server or agent settings, outside browser bundles and source control.

Whose vendor accounts and balance are used?

You receive a Dreamtune workspace and key. Dreamtune manages upstream service access, which can use shared upstream accounts. It does not create a dedicated vendor account for every customer or give you vendor login credentials. Your Dreamtune wallet, keys and run history belong to your workspace.

Supported usage made through Dreamtune consumes that wallet. Your existing provider subscriptions and balances stay separate. Dreamtune does not collect those subscriptions into a bill or transfer the credit you already hold elsewhere.

Check current pricing and offers before funding. Eligible verified accounts can claim the advertised welcome credit in Usage & billing under its stated terms. Each model, tool and GPU has its own usage rate; a shared balance does not mean unlimited usage.

Give the agent the access its task needs

A research assistant may need model and tool access without compute permissions. Create a separate named key for a project so you can review its consumption. Key spending limits control admission of new work; they are not guaranteed per-task spending ceilings.

GPU jobs need particular care: consumed runtime is billable, and ending a client request does not stop the machine. Follow the compute guide for access and shutdown.

Once the first workflow works, add another capability only when a real task needs it. For the next step, read how to make your AI agent better at real work.