How to Approve AI Tool Calls from Slack (MCP Approval Alerts)
Human-in-the-loop approval has a weakness nobody mentions until they’ve lived it: latency. You put the risky tools behind approval — good — and now an agent’s create_pull_request sits in a queue while the one person who could approve it has no idea it’s there. The agent waits, the user waits, and by the time someone opens the dashboard, the approval that should have taken thirty seconds took three hours. Do that a few times and your team starts asking to turn the approvals off — which is how safety controls die.
The fix is meeting reviewers where they already are. MCP Trail posts approval alerts to Slack: the moment a tool call parks in the approval queue, a message with the context and an approval link lands in your channel. Review happens in seconds, from the tool your team watches all day.
What you’ll need
- A free MCP Trail account, on a Team plan or above (Slack approval alerts are a team feature — the approval queue itself works on every plan)
- A Slack workspace you can authorize
- At least one tool set to human approval (how to do that)
Step 1: Put a risky tool behind approval
If you haven’t yet: open your server under Guardian servers and set the tools that deserve a human gate — outbound messages, deploys, deletes — to human approval. Calls to them now pause in the Approval queue.
Step 2: Connect Slack
Open the server’s Security tab → Approval alerts (Slack). Connect your workspace — it’s a one-click OAuth authorization, same pattern as everything else on the platform — and choose the channel that should receive approval requests.
Step 3: Trigger a test approval
Ask your assistant to call the gated tool. Watch the channel: a message arrives naming the server, the tool, and a link to review the request — while the agent’s call waits.
Step 4: Review and decide
Follow the link to see the full request — tool, arguments, requesting session — and approve or reject. Approve, and the parked call proceeds; the agent continues as if the pause never happened. Reject, and the agent gets an explicit denial it can relay. Either way the decision, the decider, and the timestamp are in the Audit log.
Making approvals sustainable
Alerting is half of it; the other half is not drowning reviewers:
- Gate narrowly. Approval on everything trains people to rubber-stamp. Reserve it for the calls a human genuinely should judge; let policies, sequence rules, and risk scoring auto-handle the rest.
- Route to a channel the right people actually read — a dedicated
#ai-approvalsbeats a busy team channel. - Watch the queue’s latency. If approvals routinely wait hours, either the channel is wrong or the gated set is too broad.
FAQ
Do approvals work without Slack?
Yes — the Approval queue in the dashboard is always there (plus standalone approval links). Slack alerts remove the “nobody saw it” delay.
Who can approve?
Reviewers with access to your MCP Trail workspace and the approval — the Slack message links into the authenticated review flow, so a random channel member can’t approve by reacting to a message.
What does the agent experience while waiting?
The tool call is held open as pending. On approval it completes normally; on rejection the agent receives an explicit policy denial rather than a mystery timeout.
Which plan do I need?
Slack approval alerts are available on Team and above. Everything else in this post — per-tool approval policies and the queue itself — works on the free tier.
Related Articles
- MCP Approvals: Human-in-the-Loop for AI Tool Calls
- How to Control Which MCP Tools Your AI Can Run
- How to Set Up the Slack MCP Server
- MCP Session Risk Scoring
- How to Enforce Tool Order in MCP Sessions
An approval queue nobody watches is just a slower “no”. Create your free MCP Trail account and wire the queue into Slack.