Claude Managed Agents Starter
https://github.com/vercel-labs/claude-managed-agents-starter- Category
- AI Agents
- Rank
- No. 1485Tools index
Previous survey · No. 1491 ·
- Pricing
- Open Source
- Type
- TOOL
- Builder
- vercel-labs
- GitHub
- 66 stars
- Date
About
Open-source template for building Claude Managed Agents on Vercel.
What it does
A deployable internal knowledge assistant that lets signed-in users question connected GitHub, Notion, and Slack sources. Claude Managed Agents handles agent work, MCP supplies source access, and Postgres stores user, connection, and agent-session data.
Why it's ranked here
The starter covers unusually broad application plumbing: authentication, database models, encrypted OAuth tokens, MCP credentials, agent sessions, and background processing. That makes it useful as an implementation reference, though deployment requires several hosted services and careful secret configuration.
What's good
OAuth tokens receive AES-256-GCM encryption before database storage. Each user can receive a separate Anthropic vault, while credential synchronization updates existing MCP credentials and handles creation conflicts. Authentication supports Vercel OIDC and optional GitHub OAuth, with account linking and encrypted provider tokens.
Tradeoffs
This is a complete hosted application stack, not a small agent library. It depends on Anthropic credentials, Neon, Vercel OAuth, Better Auth, and multiple secrets. Message limits live in process memory, so counters are instance-local and disappear when the process restarts.
How to use it well
Use it when building a Vercel-hosted internal assistant that searches GitHub, Notion, and Slack through MCP. Treat it as a starting architecture for identity, token custody, persistence, and agent sessions. It does not provide a general-purpose CLI or standalone library package.
Technical notes+
package.json defines a private pnpm Next.js 16 application with TypeScript, React 19, Drizzle, Neon, Better Auth, Workflow SDK, and the Anthropic SDK. lib/schema.ts models authentication, managed-agent sessions, MCP clients, and per-user MCP tokens. lib/mcp-oauth.ts performs OAuth discovery, optional dynamic client registration, PKCE, token exchange, encrypted token persistence, and connection lookup. lib/crypto.ts implements AES-256-GCM using TOKEN_ENCRYPTION_KEY. lib/vault.ts creates per-user Anthropic vaults and synchronizes static bearer credentials. lib/rate-limit.ts uses in-memory daily user and global counters.
Observed
- Primary language
- TypeScript
- Packaging
- Private pnpm application package
- Application interface
- Next.js web application with REST API routes described in the README
- Agent integration
- Claude Managed Agents through the Anthropic SDK
- MCP connections
- GitHub, Notion, and Slack
- Data layer
- Neon Postgres with Drizzle ORM
- Authentication
- Better Auth with Vercel OIDC and optional GitHub OAuth
- Deployment surface
- Vercel deployment flow with Neon integration
Read from README.md, package.json, lib/db.ts, lib/auth.ts, lib/time.ts, lib/utils.ts, lib/vault.ts, lib/crypto.ts, lib/schema.ts, lib/session.ts, lib/branches.ts, lib/anthropic.ts, lib/mcp-oauth.ts, lib/rate-limit.ts, lib/auth-client.ts.
What it can do
Deploy Claude managed agents
Agent configuration and code → Live deployed agent on Vercel
Create conversational AI agents
Natural language prompts and configuration → Interactive AI agent responses
Generate boilerplate agent code
Template selection and parameters → Structured agent application code
Configure agent behavior and personality
Behavioral parameters and instructions → Customized agent with specific traits
Integrate Claude API functionality
API keys and endpoint configurations → Connected Claude-powered agent
Set up development environment
Project requirements and dependencies → Ready-to-use development workspace
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.