Managed Deep Agents Skill for Claude Code: Build and Deploy an Agent on LangSmith

Walks Claude through interviewing you, scaffolding, testing and deploying a Managed Deep Agent on LangSmith with the mda CLI.

Installs: 9.6K; on skills.sh
Stars: 1.3K; on GitHub
Updated: Oct 2026; last repo commit

When you say something like "help me build an agent", Claude first asks what the agent should do and who talks to it. It checks the answer against the platform's limits, confirms the shape with you, scaffolds with mda init, adds one capability at a time, runs it with mda dev, then deploys with mda deploy.

Managed Deep Agents is in public beta on US LangSmith Cloud only, and it needs a LangSmith API key with beta access plus the mda CLI. Deploying uploads your project to LangSmith. SKILL.md does not cover pricing.

Good for

Developers building a Deep Agent in Python or TypeScript to run on LangSmith
Teams that want a hosted agent without running their own server
People adding tools, MCP servers, schedules, Slack or evals to an existing Managed Deep Agent

Install managed-deep-agents

Claude Code

Copy this prompt.

Install the managed-deep-agents skill in Claude Code. Run this command:

/plugin marketplace add langchain-ai/langchain-skills
/plugin install langchain-skills@langchain-skills

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Pick one skill route, not both. Then install the mda CLI with the Python or the TypeScript command; the package you pick sets the language mda init scaffolds. The plugin also installs the skills this one points to: deep-agents-core, langgraph-cli, langchain-middleware, eval-engineering and langgraph-human-in-the-loop. The README warns the project is in early development, and says to set OPENAI_API_KEY or ANTHROPIC_API_KEY after installing, depending on the models you use.

From the project README. Source.

What's in the skill

When it loads

INVOKE THIS SKILL when building, testing, or deploying Managed Deep Agents in LangSmith. Walks a user through their first agent end to end — interviewing them about what they want to build, mapping it onto what MDA can actually do, then scaffolding and deploying it. Covers the file-based project layout; define_deep_agent / defineDeepAgent; instructions, skills, memory, identity, tools, MCP servers, connections, middleware, sandboxes, schedules, channels, and evals; the mda CLI; GitHub deployments; and Context Hub.

Sections in SKILL.md

Overview
When to use
1. Ask what they want to build
2. Check the answer against the limits
3. Map the answer onto capabilities
4. Confirm the shape before writing files
5. Scaffold and wire the smallest thing that runs
6. Handle keys without touching their secrets
7. Run it locally, then deploy
What MDA cannot do
Prerequisites
Project layout
Define the agent
Instructions
Behavior
Skills

Key facts

Author: LangChain (vendor's own repo, langchain-ai/langchain-skills)
Status: Managed Deep Agents is public beta, US LangSmith Cloud only; the repo README says it is in early development
Needs: LangSmith API key with beta access, the mda CLI, Python with uv or Node.js with npm, and model credentials
Folder: 1 file: SKILL.md
Part of plugin: langchain-skills: 22 skill folders in the repo (the README says 21)
Installs on skills.sh: 9,584 (Nov 2, 2026)
GitHub stars: 1,320 (Nov 2, 2026)
License: MIT
Last repo commit: Oct 29, 2026

What it includes

Asks what the agent should do and who or what talks to it, then only follow-ups.
Follow-ups cover memory, private APIs, human approval and file or code access. It stops once it can name the capabilities.
1. Ask what to build
Checks the request against 9 listed limits before promising anything.
Examples: US LangSmith Cloud only, Slack as the only channel, memory shared by all callers, builds capped at 200 MB. Custom routes or non-US hosting send you to langgraph deploy.
2. Check against the limits
Maps your answer to files: instructions, tools, memory, schedules, channels, sandbox.
A 15-row table links what you describe to what to build and where it lives in the project.
3. Map to capabilities
States the plan back, with model and only the capabilities it will create.
It waits for your agreement before writing files.
4. Confirm the shape
Creates the project, runs uv sync, then adds one capability at a time.
mda init skips the sandbox with --no-sandbox. It does not create directories the plan did not call for.
Runs code · 5. Scaffold with mda init
Fills in key names in .env and leaves you to paste the values.
SKILL.md says not to write live credentials into .env, echo them, or copy a key from another project.
Write · 6. Handle keys
Compiles, opens LangSmith Studio and hot reloads for a local test.
You send a message in Studio and confirm the agent calls its tool before deploying.
Runs code · 7. mda dev
Syncs Context Hub, uploads the project, waits for DEPLOYED and sets up schedules.
Non-reserved .env entries go to the deployment as hosted secrets. --no-wait skips schedule setup.
Runs code · 8. mda deploy
Creates, lists and removes workspace secrets and OAuth registrations.
Used for MCP servers and APIs that need a secret or sign-in.
Try: “Help me build an agent that posts a weekday summary of open bugs to Slack.”
Runs code · mda connections
Deletes a deployment and the LangSmith resources it created.
SKILL.md says to confirm with you first and never pass --yes unprompted.
Delete · mda delete

From the official docs

Permissions

Medium risk

Rated medium because deploying uploads your project and forwards .env entries as hosted secrets, and mda delete removes a deployment. Check each command before you approve it and keep real secrets out of the chat.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

Deploying sends your project and secrets to LangSmith
The project source goes up as a build archive, and .env entries other than reserved ones become hosted secrets. .env itself is never archived. Review what is in .env first. github.com
mda delete removes a deployment
It deletes the deployment and the LangSmith resources it created, including managed sandboxes. SKILL.md says to confirm first and never pass --yes unprompted. github.com
Memory is shared by every caller
The memory tree is shared across all callers and any of them can influence it. Keep personal data and credentials out and treat its contents as untrusted. github.com
A global prerelease CLI is installed
The prerequisites use uv tool install --prerelease allow managed-deepagents, or npm install -g managed-deepagents@dev. The skill says the SDK moves quickly, so check flags with mda --help. github.com
Schedules and Slack run without you
Deployed schedules run on a cron and the Slack channel starts runs from messages. Check what the agent can do before you enable them. github.com

Technical details

Sign-in and keys	mda deploy uses LANGSMITH_API_KEY from .env or the shell; browser sign-in caches a short-lived personal token on your machine.
Scopes	Runs the mda CLI, uv and npm in your shell Creates a project and writes .env placeholders mda deploy uploads the project to LangSmith Cloud and forwards non-reserved .env entries as hosted secrets mda init scaffolds a sandbox by default (delete the sandbox/ directory to opt out); the deployed agent can then run shell commands and delete files in it

Questions

What does the managed-deep-agents skill do?

It guides Claude through building a Deep Agent that runs on LangSmith's managed hosting. Claude interviews you, checks what the platform can do, scaffolds the project, and tests and deploys it with the mda CLI.

What do I need before using it?

Access to the Managed Deep Agents public beta and a LangSmith API key for it, Python with uv or Node.js with npm, and credentials for your chosen model. Install the mda CLI first, as the install box shows.

How do I install it?

Use the install box above. In Claude Code, the plugin route installs all 22 skills in the LangChain skills repo, including the other skills this one points to. The single-skill command installs only this skill. There is no Claude app tab because the skill runs the mda CLI.

Does it cost money?

SKILL.md does not give pricing. Deploying runs your agent on LangSmith, and the model you choose needs its own provider key, so check LangSmith's and your provider's terms.

Is it safe to use?

It uploads your project to LangSmith when you deploy, forwards .env entries as hosted secrets, and can delete a deployment. Check each command Claude proposes, keep secrets out of the chat, and never let it pass --yes to mda delete without asking you.

Is this an official LangChain skill?

It is in LangChain's own repo, langchain-ai/langchain-skills, and the repo's plugin manifest names LangChain as author. The repo README says the project is in early development and its content may change.

When should I use langgraph deploy instead?

SKILL.md says to use a standard LangSmith Deployment instead when you need custom server routes, stronger isolation, maximum scalability or a region other than US.

Is memory private to each user?

No. SKILL.md says durable memory is one tree shared by every caller of the deployment, so never store personal data, credentials or tokens in it.

Information

updated Nov 2, 2026

Repository: langchain-ai/langchain-skills
Folder: config/skills/managed-deep-agents

langchain-ai/langchain-skills skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 3

managed-deep-agents SKILL.md · github.com
langchain-skills README · github.com
langchain-skills plugin.json · github.com