Try it hosted

30-second trial · no install

Pick your stack, get a pre-configured workspace

We mint a time-boxed DashClaw workspace and hand back a drop-in config for your agent stack. No account, no credit card — the trial runs for 30 days or 10,000 governed actions.

Connect an agent

Point your agent at your running DashClaw instance.

Pick an integration surface, configure an approval surface, verify with one command. Works with any agent framework that can call an HTTP API or an SDK.

Before you start, you need a running DashClaw instance. If you do not have one yet, stand it up first. Takes about 10 minutes on Vercel and Neon free tiers.

Self host the runtime Explore the live demo

First connection takes 5 to 15 minutes depending on the integration surface you pick.

Get your API key

From your DashClaw instance settings, copy your API key. It starts with oc_live_ for self hosted instances. Export it alongside your instance URL:

export DASHCLAW_BASE_URL=https://your-instance.vercel.app
export DASHCLAW_API_KEY=oc_live_...

Never use https://dashclaw.io as your agent base URL. Point at your own instance. The dashclaw.io deployment runs in demo mode and rejects writes.

Pick an integration surface

Five ways to plug an agent into DashClaw. Pick one. All five hit the same governance loop on the same instance, so you can switch later without changing anything else.

SDK

Node.js or Python. Wrap risky actions in claw.guard() and you are done.

# Node
npm install dashclaw

# Python
pip install dashclaw

Full SDK docs

MCP Server

Zero code. Any MCP compatible client (Claude Code, Claude Desktop, Claude Managed Agents) gets governance through one config block.

npx @dashclaw/mcp-server \
  --url $DASHCLAW_BASE_URL \
  --key $DASHCLAW_API_KEY

MCP server docs

Claude Code Hooks

Two Python files dropped into .claude/hooks/. Governs Bash, Edit, Write, MultiEdit, sub-agent spawns, and MCP tool calls (mcp__*) such as Gmail/Stripe/Calendar sends. Safe to ship even without DashClaw configured.

cp hooks/dashclaw_*.py .claude/hooks/

Claude Code guide

OpenClaw Plugin

Framework native plugin for OpenClaw agents. Intercepts PreToolUse and PostToolUse, runs guard, records the outcome, and waits for approval automatically.

npm install @dashclaw/openclaw-plugin

OpenClaw plugin guide

Custom connector (Claude app — web / Desktop), OAuth, no key

No API key. Paste your instance URL into Claude's Add custom connector, then log in and authorize. Free tier allows one connector. On plain chat, governance is advisory — it records actions and prompts for approval, not a hard block.

https://YOUR-INSTANCE.vercel.app/api/mcp

In Claude, open Add custom connector, paste the URL, click Connect, log in, and Authorize.

Claude app connector guide

Pick an approval surface

When guard returns require_approval, the action pauses until a human resolves it. Pick where humans should see and resolve those approvals. Dashboard is on by default. The other four are optional and additive.

Dashboard

Default

Always on. Interactive queue with triggering policy, risk score, and replay link.

https://<your-instance>/approvals

CLI

For terminal first developers. dashclaw approve and dashclaw deny against the same governance endpoint.

npm install -g @dashclaw/cli
dashclaw approvals

Mobile PWA

Add /approve to your home screen on iOS or Android. One tap Allow or Deny from the phone. SSE driven, updates within about one second.

https://<your-instance>/approve

Discord bot

Phone first via DM. Inline Approve and Deny buttons in a registered user DM. Fire and forget; action creation succeeds even if Discord is unreachable.

DISCORD_BOT_TOKEN=<bot-token>
DISCORD_APPROVER_USER_ID=<your-discord-user-id>
DISCORD_APPROVER_ORG_ID=<your-org-id>
DISCORD_PUBLIC_KEY=<discord-app-public-key>

Also set the Interactions Endpoint URL in the Discord Developer Portal to https://<your-instance>/api/discord/interactions.

Discord approvals guide

Telegram bot

Inline Approve and Deny buttons pushed to an admin chat. Warns and moves on if Telegram is unreachable.

npm run telegram:setup

Telegram setup guide

Verify

Run dashclaw doctor from any terminal. Exit 0 means the instance is healthy and your SDK or surface is reachable. Then have your agent attempt a low risk action and watch it land in the dashboard inbox.

Expected proof: dashclaw doctor exits 0 or names the blocker, /decisions shows the action record, /approvals shows any held action, and /api/setup/live-proof can capture setup evidence for onboarding or CI.

From any terminal

dashclaw doctor

Diagnoses database, configuration, auth, deployment, SDK reachability, governance staleness, and shape drift. Auto fixes safe issues.

npm install -g @dashclaw/cli
dashclaw doctor

Self host operator

npm run doctor

Adds filesystem level fixes (env writes, migrations, default policy seed). Backs up .env before any write.

npm run doctor

Exit codes: 0 healthy, 1 warnings or unreachable. Add --json for CI integration.

If it doesn't connect

· The MCP server reads DASHCLAW_URL while the SDK and hooks read DASHCLAW_BASE_URL — set both to your instance URL.
· 401 Invalid or missing API key right after a deploy or update usually means the DB schema is behind — run npm run db:migrate on the host.
· Never point an agent at the demo deployment (e.g. dashclaw.io) — it rejects writes. Use your own instance.
· Hook not firing? Re-run npm run hooks:install and confirm the matcher in .claude/settings.json includes mcp__.*.

Framework guides

Step by step walkthroughs for popular frameworks

Deeper walkthroughs once you have picked a surface in Step 2. Each takes 10 to 20 minutes end to end.