curl -LsSf https://langch.in/dcode | bash
dcode

uv tool install deepagents-cli

You'll need a LangSmith API key with access to the Managed Deep Agents private preview (waitlist). Export it before running any command, or put it in a repo .env or ~/.deepagents/.env:

export LANGSMITH_API_KEY="..."

# Scaffold a new project folder
deepagents init my-agent

# Register any MCP servers you intend to use (one-time, per workspace)
deepagents mcp-servers add --url https://tools.langchain.com \
                            --header X-Api-Key=$LANGSMITH_API_KEY \
                            --name Fleet

# init scaffolds an empty tools.json. Add tools that reference a registered
# MCP server's URL (otherwise the agent has no tools), e.g. the Fleet server
# registered above:
#   { "name": "read_url_content", "mcp_server_url": "https://tools.langchain.com",
#     "mcp_server_name": "Fleet", "display_name": "read_url_content" }

# Upsert the project as a managed agent on /v1/deepagents/*
cd my-agent && deepagents deploy

deepagents init scaffolds an empty tools.json, an example skill under skills/, and an example subagent under subagents/; edit or remove them to suit your agent. tools.json starts empty because every tool must reference an MCP server that is already registered in the workspace (deepagents mcp-servers add) — so a freshly scaffolded project deploys without first registering a server.

New agents default to the state backend. To opt into a managed sandbox, set agent.json's backend.type to sandbox and configure backend.sandbox_config.scope as thread or agent; sandbox_config can also include sandbox policy IDs and TTL fields. The CLI does not create or run sandboxes locally; sandbox lifecycle is handled by the Managed Deep Agents platform.

{
  "backend": {
    "type": "sandbox",
    "sandbox_config": {
      "scope": "thread",
      "policy_ids": ["policy-id"]
    }
  }
}

my-agent/
  agent.json              # name, description, backend, runtime.model, permissions
  AGENTS.md               # system prompt
  tools.json              # tools the agent can call (optional)
  skills/<name>/SKILL.md  # frontmatter-tagged skills (optional)
  subagents/<name>/       # delegated subagent definitions (optional)

deepagents agents list                  # list workspace agents
deepagents agents get <agent_id>        # show one agent
deepagents agents delete <agent_id>     # delete an agent

deepagents mcp-servers list                  # list workspace MCP servers
deepagents mcp-servers add --url URL          # register a server
deepagents mcp-servers get <id|name|url>      # show one server
deepagents mcp-servers tools <id|name|url>    # list a server's tools
deepagents mcp-servers update <id|name|url>   # update server URL or headers
deepagents mcp-servers delete <id|name|url>   # remove a server
deepagents mcp-servers connect <id|name|url>  # start OAuth for a server

get, update, delete, and connect accept an MCP server's id, exact name, or URL — a non-id value is resolved against mcp-servers list (URLs are matched ignoring case and trailing slash).

• CLI Documentation
• Changelog
• Source code
• Deep Agents SDK — underlying agent harness
• Deep Agents Code — coding agent
• LangChain Academy — Comprehensive, free courses on LangChain libraries and products, made by the LangChain team.
• Code of Conduct — community guidelines and standards

See our Releases and Versioning policies.

As an open-source project in a rapidly developing field, we are extremely open to contributions, whether it be in the form of a new feature, improved infrastructure, or better documentation.

For detailed information on how to contribute, see the Contributing Guide.