Claude Research Analyst for Slack

Claude Research Analyst for Slack is a boilerplate that turns any Slack workspace into a persistent AI research analyst, powered by Claude Managed Agents, Chat SDK, and Vercel Connect.

What is the Claude Research Analyst for Slack?

It is a starter template from Vercel Labs that integrates a Claude Managed Agent with Slack through Chat SDK. When a user @mentions the bot in a channel or sends it a DM, the agent searches and fetches sources inside an Anthropic-managed sandbox and streams a brief back into the thread. The app runs on Next.js 16 on Vercel and uses Claude Sonnet 5 as its underlying model.

Each Slack thread maps to one persistent Managed Agents session, so follow-ups keep the prior research context. Vercel Connect owns the Slack app and credentials, while Upstash Redis stores subscriptions, session mappings, and webhook deduplication across deployments.

Key Features

• Persistent thread context — Every Slack thread maps to one Managed Agents session; follow-ups retain prior research context without re-mentioning in single-human threads.
• One-command provisioning — npm run cma:setup creates the agent and sandbox environment, writes the agent ID and environment ID to .env.local, and adds them to the linked Vercel project.
• Anthropic-managed sandbox — The agent searches and fetches web pages in a sandbox where Bash is disabled to reduce the risk of untrusted pages exfiltrating conversation data.
• Debug mode diagnostics — Setting CLAUDE_DEBUG_MODE=true posts a table after each turn with duration, model requests, token and prompt-cache usage, web searches and fetches, and a Claude Console session link.
• Participant-aware mentions — Follow-ups in a thread need no extra @mention when one human participates; with multiple humans, the bot responds only when explicitly mentioned.
• Chat SDK Slack adapter — Slack events are acknowledged quickly by the webhook route and dispatched in the background, using the official Chat SDK Slack adapter.
• Redis-backed state — Upstash Redis preserves subscriptions, thread/session mappings, and webhook deduplication so restarts and redeploys don't lose context.
• Customizable model and prompt — The model and system prompt live in scripts/cma/lib/agent.ts; publishing a new version via npm run cma:update is required after prompt changes.

Who is it for?

• Teams that want an AI researcher inside Slack — Deploy it to a channel and get cited, sourced briefs without leaving the conversation.
• Developers building on Claude Managed Agents — Use the repo as reference for wiring managed agents to Slack with Chat SDK, Vercel Connect, and Redis.
• Vercel and Upstash users — See an example of Vercel Connect Slack triggers, marketplace Redis, and serverless webhooks working together.

What can you do with it?

• Research in channels: Mention the bot in a thread and it will pull live web sources, then stream a brief reply directly back into the thread.
• Persistent follow-up questions: Ask a follow-up in a single-human thread without re-mentioning; context carries over from earlier turns in the same thread.
• Debug agent behavior: Turn on CLAUDE_DEBUG_MODE=true to see per-turn diagnostics, token usage, and a link to the Claude Console session for inspection.

How does it work?

Setup follows four steps: clone the repo and install dependencies (Node.js 24+), attach Slack via Vercel Connect and add Upstash Redis from the Vercel Marketplace, run npm run cma:setup to provision the Managed Agent and sandbox, then add ANTHROPIC_API_KEY and deploy with vercel deploy --prod. At runtime, Slack webhooks are received by the Next.js route, mapped to a persistent agent session, and the event stream is streamed back into the thread as the agent works.

Pros and cons

• Pros: one-command setup, persistent per-thread context, structured debug output, and fast webhook acknowledgement via background dispatch.
• Cons / considerations: requires Node.js 24+, access to Claude Managed Agents (Anthropic API key), a Redis instance, and all three required environment IDs; archiving resources with npm run cma:archive is permanent, so confirm agent IDs first.

FAQ

What do I need before deploying this template?

You need an Anthropic API key with Managed Agents access, a Vercel project, and a Redis store (Upstash Redis from the Vercel Marketplace is used in the quickstart). The setup also requires Node.js 24+ locally to run the provisioning script.

Does the bot remember context from earlier turns?

Yes. Each Slack thread maps to one persistent Managed Agents session, so follow-ups in that thread retain the prior research context. Anthropic pins each session to the agent version that created it, so prompt changes should be tested in a new thread.

How do I change the model or system prompt?

Edit scripts/cma/lib/agent.ts and publish the change with npm run cma:update. Do not rerun npm run cma:setup for prompt changes; that command is only for one-time provisioning.

How do I clean up the resources?

Run npm run cma:archive to archive the persistent Anthropic resources. Because archiving is permanent, confirm both Managed Agents IDs before executing.