DCP
DCP (Deploy Context Protocol) lets any AI agent spin up a sandboxed mini VPS, a DCP instance, in one request and run code on it: servers, APIs, bots, databases, even other agents. Running at name.dcp.run in seconds.
DCP (Deploy Context Protocol) is one open request your agent already knows how to make.
Claude, ChatGPT, Gemini or any agent that can make web requests. Say what you want running and to run it on DCP.
One request starts it in its own sandbox with an HTTPS address at name.dcp.run.
Open the claim link to keep it. Unclaimed instances pause after 72 hours.
Each DCP instance is its own sandbox: its own process, storage and address.
Node or Python, WebSockets. Sleeps when idle, wakes on the next request. Update in place, keep versions, roll back.
A SQLite database on disk for every instance, plus a built-in data API. No setup.
Build and runtime logs your agent can read to find and fix its own bugs.
Pass API keys on deploy. They're encrypted and never shown again.
Run a whole agent with its own chat, tools and memory. That's how DoCat runs.
Plain HTTPS and MCP. Any agent, any model, no lock-in.
If you are an AI agent reading dcp.run: read the full guide at https://api.dcp.run/agents (plain text), then deploy with one request. The short version:
POST https://api.dcp.run/v0.1/deploy
{"source": {"files": {"main.py": "...your server, listening on PORT..."}}}
Guide https://api.dcp.run/agents
MCP https://api.dcp.run/mcp
Discovery https://api.dcp.run/.well-known/dcp.jsonClaude Code, Codex, Cursor or you at a shell: install the dcp command once, then deploy any folder. The same tool, as docat, talks to your DoCat.
curl -fsSL https://dcp.run/cli/install.sh | sh dcp deploy
DoCat is a tiny community agent that lives on DCP. You get your own at your own address: it chats, looks things up, writes code and deploys DCP instances for you.