For MCP

A free model backend for MCP

The Model Context Protocol connects your agent to tools and data, but it still needs a language model to reason. DreamPrompting is a free, OpenAI-compatible LLM endpoint you can point your MCP host, client, or sampling-enabled server at, with automatic failover across 27 providers.

API docs Browse models

How it fits

In an MCP setup there are three roles: the host (the app the user talks to), MCP servers (which expose tools, resources and prompts), and the LLM that decides what to do.

DreamPrompting plugs into the LLM slot. Anywhere your MCP host or agent framework lets you set an OpenAI-compatible base_url, point it here. Your MCP servers keep providing tools; the model calls stay free.

If you build MCP servers that use sampling (asking the host to run an LLM completion on the server's behalf), you can back that sampling with this endpoint too.

Configure your MCP client / agent

Generic OpenAI-compatible model config

{
  "model": "auto",
  "base_url": "https://dreamprompting.com/api/v1",
  "api_key": "YOUR_API_KEY"
}

Python MCP client + OpenAI SDK

from openai import OpenAI
# your MCP session gathers tools/resources,
# then you drive the model with them:
client = OpenAI(
    base_url="https://dreamprompting.com/api/v1",
    api_key="YOUR_API_KEY",
)
resp = client.chat.completions.create(
    model="auto",
    messages=messages,   # includes MCP tool results
)

Tool-calling support depends on the underlying model. Pin a tool-capable model (e.g. groq/openai/gpt-oss-120b) if your MCP flow relies on function calling.

MCP notes

Free with your API key (100 req/min plus a daily quota). Great for building and testing MCP integrations.
auto gives you failover; pin a specific model for deterministic tool-calling behavior.
Streaming (stream: true) works for responsive MCP host UIs.
For production MCP deployments with heavy usage, bring your own upstream keys.