Kiro Not Working? 7 Common Causes and How to Fix Them (2026)

When Kiro stops responding, the most likely causes are a monthly vibe/spec request limit being hit, an expired AWS Builder ID session, or a stalled MCP server connection, not a full outage. This guide is for developers using Kiro's agentic IDE who are stuck mid-task. The fix usually takes under 5 minutes: 3 of the 7 causes below are permission or auth resets, 2 are quota-related, and the rest are local app issues like cache or extension conflicts.

Why is Kiro not responding?

Kiro usually stops responding because it's waiting on an MCP (Model Context Protocol) server that hung, or because the agent loop in Autopilot mode got stuck in a retry cycle. Restarting the Kiro window (Cmd/Ctrl+Shift+P → "Reload Window") clears most stalls without losing your spec or task history.

If reloading doesn't help, check whether you're in Autopilot or Supervised mode. Autopilot can loop on a failing test or build step if it keeps retrying the same fix. Switching to Supervised mode for that task lets you approve each step manually and break the loop.

Is Kiro down right now?

Kiro is built on AWS infrastructure, so regional AWS incidents (particularly in us-east-1, which hosts many Bedrock-backed services) can affect it even when your local setup is fine. Check the AWS Service Health Dashboard and Kiro's own status/release notes at kiro.dev before assuming it's a local bug.

A quick way to tell: if other AWS-dependent tools (Bedrock console, Amazon Q) are also degraded, it's an AWS-side issue and you just need to wait it out.

Why am I getting a rate limit or quota error?

You've hit your monthly vibe request or spec request allowance for your Kiro plan. Kiro meters usage separately for conversational "vibe" requests and structured "spec" requests, and once either pool is exhausted, further requests return a limit error until your plan resets or you upgrade.

For a full breakdown of how these pools work, current plan quotas, and what triggers a hard stop versus throttling, see our guides on Kiro pricing and rate limits and what to do when Kiro's rate limit is exceeded.

How do I fix Kiro authentication errors?

Sign out and back in through your AWS Builder ID or IAM Identity Center connection ("Kiro: Sign Out" in the command palette, then sign back in). Session tokens expire periodically, and an expired token is the most common cause of silent failures where Kiro accepts input but never returns a response.

• Confirm your system clock is accurate (clock drift breaks token validation)
• If you're on a corporate IAM Identity Center setup, confirm your admin hasn't revoked or rotated access
• Clear the Kiro extension's cached credentials folder if re-signing in doesn't resolve it

Why can't Kiro connect to my MCP servers?

MCP server connection failures are usually a local network, port, or config issue, not a Kiro bug. Check your mcp.json or workspace settings for a malformed server command, a port already in use, or a firewall blocking the local socket Kiro uses to talk to the server process.

Run the MCP server command manually in a terminal first. If it errors outside of Kiro, the problem is in your server setup, not the IDE.

What's the fastest way to troubleshoot Kiro?

Work through these steps in order before filing a support ticket; most "not working" reports resolve within the first three steps.

Step	Action	Fixes
1	Reload window	Stuck UI, frozen chat panel
2	Check AWS status page	Regional outages, Bedrock degradation
3	Check usage/quota in settings	Vibe/spec limit reached
4	Sign out / sign back in	Expired auth token
5	Test MCP server command directly in terminal	Tool connection errors
6	Disable other VS Code extensions temporarily	Extension conflicts
7	Update Kiro to the latest release	Known bugs already patched

Should I switch tools if Kiro keeps failing?

If the failures trace back to request quotas rather than bugs, switching tools won't necessarily help since most AI IDEs (Cursor, Claude Code, Windsurf) meter usage similarly. It's worth comparing how each handles limits before you commit to a switch; our Kiro vs Cursor, Kiro vs Windsurf, and Kiro vs Claude Code comparisons break down quota structure, pricing, and workflow differences side by side.

If quota exhaustion is the recurring issue rather than outright bugs, upgrading your Kiro plan or spacing out large spec runs across the month is usually faster than migrating your whole workflow to a different IDE.

Key takeaways

• Most "Kiro not working" reports are quota, auth, or MCP config issues, not core product outages.
• Check the AWS Service Health Dashboard before assuming it's a Kiro-specific bug, since Kiro depends on AWS/Bedrock infrastructure.
• Vibe requests and spec requests are metered separately, so you can exhaust one pool while the other still has headroom.
• Reloading the window and re-authenticating resolve the majority of stalls without losing in-progress specs or tasks.