DreamAgent MCP Server

Standalone MCP server that lets ChatGPT operate your DreamAgent account: create projects (Telegram bots, Discord bots, websites, schedulers), watch them build, and edit them conversationally through DreamAgent's AI agent.

It talks only to the existing DreamAgent REST API — no backend changes, and all server-side ownership checks apply unchanged.

Tools

Tool	What it does
dreamagent_list_projects	List your projects (id, type, status, domain)
dreamagent_create_project	Create a telegrambot / discordbot / website / scheduler (async)
dreamagent_get_project_status	Poll a project until ready / failed
dreamagent_chat	Send a build/edit/debug instruction to the project's AI agent (async, consumes credits)
dreamagent_get_chat_status	Poll the agent run; streams new output + next_after cursor
dreamagent_cancel_chat	Cancel a running edit

Long operations are submit-then-poll: create → poll status; chat → poll chat status. The tool descriptions tell ChatGPT to do this.

Setup

cd mcp-server
pip install -r requirements.txt
cp .env.example .env       # fill in your DreamAgent email + password

The account is connected by logging into the existing POST /auth/login endpoint; the session token is cached in memory and refreshed on expiry.

Run

# stdio (local clients: Claude Desktop, Claude Code, etc.)
python server.py

# HTTP (remote MCP — for ChatGPT apps)
python server.py --http --port 8800
# then expose https://your-host:8800/mcp via your reverse proxy

Connect to ChatGPT (developer apps / MCP)

1. Run the server with --http behind HTTPS (e.g. https://mcp.dreamagent.cloud/mcp).
2. In your ChatGPT app config (or any MCP client that supports remote servers), point the MCP endpoint at that URL.
3. Per-user credentials: set DREAMAGENT_EMAIL / DREAMAGENT_PASSWORD for the account the server should act as. (For a multi-user deployment, run one instance per user or wait for the OAuth phase.)

Try it locally (smoke test)

python - <<'PY'
from server import mcp
tools = mcp._tool_manager.list_tools() if hasattr(mcp, "_tool_manager") else []
print([t.name for t in tools])
PY
# or simply:
fastmcp dev server.py

Example ChatGPT conversations

• "Show my DreamAgent projects"
• "Create a Telegram bot called WeatherBot, token 123456:ABC — it should reply with the weather when I send a city name" → create → poll until ready
• "Add a /price command to my Discord bot" → dreamagent_chat → poll until done
• "Is my website finished building?"
• "Cancel the edit running on my crypto bot"

Security notes

• Credentials live only in the .env you control; the session token is transient and refreshed via the existing login endpoint.
• Every API call carries the user's own token — DreamAgent's per-user authorization and rate limits apply exactly as in the dashboard.
• Bot tokens passed to dreamagent_create_project are sent to DreamAgent over HTTPS and stored in the project's .env, the same as creating via the dashboard.