DreamAgent MCP Server
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@DreamAgent MCP ServerCreate a Telegram bot called WeatherBot that replies with the weather."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
DreamAgent MCP
Turn natural-language requests into deployed applications.
Connect ChatGPT, Claude, or any MCP client to DreamAgent — build websites, Telegram bots, Discord bots and schedulers from a conversation, watch them go live, then keep iterating.
streamable-http · OAuth 2.1 + PKCE · 24 tools · Python / FastMCP
What DreamAgent does
DreamAgent is an AI app-builder platform: you describe what you want, its agent writes the code, runs the tests, rebuilds, redeploys to production and commits to git — automatically. This MCP server exposes that power to your AI chat, so:
You: "Create a Telegram bot called WeatherBot — token
123:ABC— it replies with the weather when I send a city name."ChatGPT: creates it → polls → "WeatherBot is live."
You: "Add a /forecast command for 5-day forecasts."
ChatGPT: sends the edit → the agent codes, tests, redeploys → done.
Related MCP server: ChatGPT Local Coder
Tools
Tool | What it does |
| List your projects (id, type, status, domain) |
| Create a website / Telegram bot / Discord bot / agent project |
| Poll a build until |
| Send a build/edit/debug instruction to the project's AI agent |
| Poll an edit run (incremental output) |
| Edit progress by project — no session key needed |
| List conversation threads per project |
| Start a fresh topic thread |
| Cancel a running edit |
| Last-resort unlock (confirm first) |
| List the project file tree |
| Read one project source file |
| Targeted find → replace edits to an existing file (preferred for edits) |
| Create / overwrite a file (confirm-gated, new files & full rewrites) |
| Pending changes vs last commit |
| Delete a file (confirm-gated) |
| Commit + push all pending changes as one version (or retry a failed commit) |
| Version history (hash, message, status) — restorable in the dashboard |
| Rebuild and publish the project (confirm-gated) |
| Upload a design image → agent redesigns the UI to match it |
| Screenshot any public page (returns the image + URL) |
| The project's AI tool catalog (song, image, video tools) |
| Project env var names (values masked, secrets hidden) |
| Which providers the account owner connected |
Long operations are submit-then-poll — create → poll status, edit → poll progress. Every file change is committed and pushed automatically, so each edit lands in session history and version history (restorable). The agent handles code, tests, rebuild, redeploy and git commits by itself; prompts only describe the change.
Connect — ChatGPT
With the official app (recommended): find DreamAgent in the ChatGPT app directory → Connect → sign in with your DreamAgent account (email or Google).
As a custom connector:
In DreamAgent → Settings → Connect to ChatGPT → Create Key → copy the full URL (it contains your key)
ChatGPT → Settings → Apps & Connectors → Advanced → Add custom connector
Paste the URL as the MCP Server URL, choose OAuth (or No authentication — the key rides in the URL), save
Ask: "Show my DreamAgent projects"
Connect — Claude
Claude → Settings → Connectors → Add custom connector:
Name:
DreamAgentRemote MCP server URL:
https://mcp.dreamagent.cloud/mcpOAuth: leave empty — Claude auto-discovers it
Click Connect, sign in with DreamAgent, done.
Connect — any MCP client
{
"mcpServers": {
"dreamagent": {
"type": "streamable_http",
"url": "https://mcp.dreamagent.cloud/mcp"
}
}
}Auth is OAuth 2.1 (authorization-code + PKCE, RFC 7591 dynamic registration)
— or pass a DreamAgent API key as Authorization: Bearer da_... /
?key=da_....
Example prompts
"Show my DreamAgent projects"
"Create a Discord bot called PriceBot — token
<from Discord dev portal>— it shows crypto prices with /price and /top commands""Is my website finished building?"
"My crypto bot stopped replying — what's wrong?" (edits start with the agent reading the project logs)
"Add a dark-mode toggle to my documentation site"
"Roll back yesterday's dashboard change" (via chat)
"How many AI credits do I have left?"
How it works
ChatGPT / Claude / Glama
│ MCP (streamable HTTP) + OAuth 2.1
▼
mcp.dreamagent.cloud ── this server (FastMCP)
│ per-request Bearer = the user's own DreamAgent API key
▼
api.dreamagent.cloud ── DreamAgent platform
└── agent writes code → tests → rebuild → deploy → git commitEvery request runs as the connected user — DreamAgent's ownership checks, plan limits and AI-credit billing apply exactly as in the dashboard.
Security & privacy
Your key, your account. OAuth issues a revocable DreamAgent API key (
da_...) at connect time — visible and revocable any time in Settings → Connect to ChatGPT. Revoking disconnects that client instantly; other clients keep working.OAuth done right. Authorization-code + PKCE (S256) mandatory, single-use 120-second codes, redirect-URI allowlists (loopback per RFC 8252), login rate-limiting, secrets validated when presented. Client registrations persist across restarts.
No secrets in chat. Bot tokens for Telegram/Discord travel over HTTPS to DreamAgent and are stored server-side as project env vars — the same as the dashboard's create dialog. The MCP server never logs credentials.
Tokens are never stored by this server — it forwards the caller's own Bearer per request and keeps no database.
Privacy: dreamagent.cloud/privacy · Terms: dreamagent.cloud/terms
Screenshots
Coming soon — a GIF of "Create a Telegram bot" end-to-end from ChatGPT.Self-hosting / development
The hosted endpoint (mcp.dreamagent.cloud) is the easiest way to connect.
To run your own:
git clone https://github.com/nivethaug/dreamagent-mcp-server
cd dreamagent-mcp-server
pip install -r requirements.txt
cp .env.example .env # optional: your account for local stdio mode
python server.py # stdio
python server.py --http --port 8800 # remote MCP (put behind HTTPS)Hardening on the production host: the service runs under pm2 via
run-mcp.sh, which joins the dedicated cgroup dreamagent-mcp-egress.
Host firewall rules (/usr/local/sbin/mcp-egress-rules.sh, applied by the
mcp-egress-block.service systemd unit) block NEW connections from that
cgroup to all internal ranges (loopback/RFC1918/link-local/CGNAT/multicast,
v4+v6) while allowing DNS and public traffic — so screenshots can never
reach internal services, even through redirects or subresources.
System dependency — Chromium: the dreamagent_screenshot_website
tool renders pages with a system Chromium/Chrome binary (checked in order:
/usr/bin/chromium, /usr/bin/chromium-browser, /usr/bin/google-chrome,
then PATH). pip does not install it — on Debian/Ubuntu:
apt-get install -y chromium # or chromium-browser / google-chromeWithout it, every other tool works; the screenshot tool returns a clear "no headless chromium binary" error.
Three modes of authentication, in priority order per request:
Authorization: Bearer da_...header (OAuth clients / API clients)?key=da_...URL parameter (NoAuth custom connectors)DREAMAGENT_EMAIL/DREAMAGENT_PASSWORDenv (personal self-host)
Layout: server.py (tools) · oauth.py (OAuth 2.1 provider: /register,
/authorize, /token, /.well-known/*) · dreamagent_client.py (API
client) · auth.py (login mode) · plugin/ (Codex plugin package).
OAuth client registrations persist in oauth_clients.json; extend trusted
gateway callbacks via TRUSTED_OAUTH_REDIRECTS.
Links
DreamAgent — dreamagent.cloud
Documentation — dreamagent.cloud/documentation
Help & Support — dreamagent.cloud/help
Support — support@dreamagent.cloud
License
MIT
This server cannot be deployed
Maintenance
Related MCP Connectors
Connect any AI agent to 1,000+ apps and 27,000+ actions through one remote MCP server (OAuth).
Automate 1,000+ services from any MCP-compatible AI agent: build Applets, run actions and queries.
Remote MCP server for AI.TV creators — delegate account operations to your AI agent over MCP.
Let AI agents query data and act across all your business apps via MCP.
Related MCP Servers
- AlicenseNot gradedqualityAmaintenanceEnables ChatGPT to inspect and edit local projects through a secure MCP interface, offering workspace management, file operations, git integration, and safe command execution.5MIT
- AlicenseNot gradedqualityCmaintenanceTurns ChatGPT web into a local coding agent, enabling file edits, shell commands, Git operations, patches, and process management through 40+ MCP tools.MIT
- AlicenseNot gradedqualityAmaintenanceEnables ChatGPT to operate authorized local projects through MCP, including reading and modifying code, viewing Git changes, running configured programs, and reading local Codex sessions to continue work.63 npm6MIT
- FlicenseNot gradedqualityCmaintenanceEnables ordinary ChatGPT web Chat to work directly on a local machine over a private MCP tunnel — reading and editing project files, running shell commands and tests, loading local skills, manipulating images, and driving Chrome — without spinning up a separate coding agent or additional model-generation API calls. It also supports multi-step task plans with persisted execution receipts and acceptance checks so results from real runs are returned to the conversation.-