Complete Troubleshooting Guide: When Claude Won't Respond, Is Slow, or Stops Mid-Response

If you've been using Claude, you've likely encountered moments when it won't respond, takes forever to reply, or stops mid-sentence. These frustrating situations seem complex at first, but the good news is that the underlying causes are usually straightforward to identify and fix. This guide walks you through diagnosing and resolving these issues systematically.

Three Main Reasons Why Claude Stops Responding

When Claude isn't responding, the cause typically falls into one of three categories.

1. Anthropic's Server Is Down or Under Maintenance

Claude's infrastructure serves users worldwide, and occasionally maintenance windows or service disruptions occur. This is beyond your control but easy to verify.

How to check:

• Visit Anthropic's status page to see if any incidents are reported
• Search Twitter/X for "Anthropic status" to see if there are official announcements
• Try accessing Claude from a different browser or device to confirm the issue isn't device-specific

What to do:

• Wait for the service to restore. This is usually the fastest solution
• If you have urgent work, temporarily use an alternative AI service (Google Gemini, etc.)
• Check back in 15-30 minutes for service restoration updates

2. Browser-Level Issues (Cache, Cookies, Session Timeout)

Browsers accumulate cache and cookies over time, which can cause Claude to behave unpredictably or refuse to load.

How to check:

• Open your browser's developer tools (F12), go to Console, and look for error messages
• Try accessing Claude from a different browser (if you use Chrome, try Firefox or Edge)
• Note whether the problem is consistent across all your devices

What to fix (in order):

1. Clear your browser's cache and cookies, then refresh Claude
2. Close and restart your browser completely
3. Open Claude in an incognito or private browsing window
4. Switch to a different browser entirely

3. Account-Related Problems (Expired Session, Plan Issues)

Your API key may have expired, your account could be temporarily restricted, or there might be a billing issue preventing access.

How to check:

• Visit Claude's official website and verify you're logged in
• Check whether you see any error messages (save these details)
• Go to your account settings and confirm your plan status and billing information are current

What to do:

• Log out completely and log back in
• Reset your password and try logging in again

When Claude Is Slow: A Diagnostic Checklist

If Claude is responding but sluggishly, work through these checks in order.

Is Your Prompt Too Long?

Longer prompts take longer to process. This is especially true when:

• You're including 5+ images
• Your text exceeds 5,000 characters
• You're pasting thousands of lines of code
• You're uploading multiple PDF documents

Solution:

• Simplify your question and make it more concise
• Include only the information that's essential
• Break large files into smaller chunks and process them separately

Are You Hitting Peak Usage Times?

Claude's response times slow during peak usage hours, typically around midday (11 AM–3 PM UTC) and evening (7 PM–11 PM UTC). This varies by region.

Solution:

• Try using Claude during off-peak hours (early morning or late night)
• Schedule important tasks for quieter times
• If you use Claude's API, implement request batching

Are You Using an Older Model?

Claude has multiple versions (Claude 3.5 Sonnet, Claude 3 Opus, etc.). Older models may be slower than the latest version.

How to check:

• Look at the model name displayed in the top-left corner of your chat window
• Click the model selector to see what versions are available to you

Solution:

• Switch to the latest Claude 3.5 Sonnet model
• Disable Extended Thinking ("thinking mode") if you have it enabled, as it adds processing overhead

Is Your Internet Connection Stable?

Slow or unstable Wi-Fi can significantly impact response times. Connection dropouts are often mistaken for Claude being slow.

How to check:

• Load a website like YouTube to confirm your connection is responsive
• Run a speed test at Speedtest.net
• Check whether other devices on your network are using bandwidth

Solution:

• Move closer to your Wi-Fi router or switch to a wired connection
• Switch to mobile data (4G/5G) if available
• Restart your router

Why Claude's Response Stops Mid-Answer

When Claude's response cuts off abruptly, here are the most common causes.

You've Hit the Token Limit

Claude has a limit on how much text it can generate in a single response. For very long answers, it may stop at this boundary.

Solution:

• Type "continue" or "keep going" to prompt Claude to finish the response
• Request shorter, more focused answers (e.g., "summarize in 300 words")
• Use Claude 3.5 Sonnet, which has higher token limits than older models

Your Network Connection Dropped

If your connection cuts out while Claude is generating a response, the output stops at that moment.

How to check:

• Look for connection error messages in your browser
• Open developer tools (F12 → Network tab) to see if requests show failures
• Try the request again to see if it completes this time

Solution:

• Verify your internet connection is working
• Resubmit the same prompt
• If using a VPN, try disabling it and requesting again

Your Browser Crashed or Froze

Browser instability can cause the tab or window to become unresponsive mid-response.

How to check:

• Can you interact with other browser tabs or windows?
• Check Task Manager (Windows) or Activity Monitor (Mac) for high memory usage by your browser
• Try moving your mouse or clicking elements on the page

Solution:

• Close and fully restart your browser
• Update your browser to the latest version
• Close unnecessary tabs to free up memory
• Try a different browser

Timeout Issues with Claude Code and API

If you're using Claude Code or accessing Claude through its API, timeouts are a specific concern.

Common timeout errors:

• timeout: The server took too long to respond
• 429: You've exceeded your API rate limit

Code example: Handling timeouts in API calls

import anthropic
import time
 
client = anthropic.Anthropic(api_key="YOUR_API_KEY")
 
# Implement retry logic for timeout resilience
for attempt in range(3):
    try:
        message = client.messages.create(
            model="claude-3-5-sonnet-20241022",
            max_tokens=1024,
            messages=[
                {"role": "user", "content": "Process this large dataset"}
            ],
            timeout=60.0  # Set timeout to 60 seconds
        )
        print(message.content[0].text)
        break
    except anthropic.APIStatusError as e:
        if e.status_code == 429:
            # Rate limited; wait and retry
            wait_time = 2 ** attempt
            print(f"Rate limited. Waiting {wait_time} seconds...")
            time.sleep(wait_time)
        elif e.status_code == 408:
            # Timeout; retry with backoff
            print(f"Timeout on attempt {attempt + 1}. Retrying...")
            time.sleep(2 ** attempt)
        else:
            raise

How to fix timeout errors:

1. For timeout errors: Break your request into smaller pieces
2. For rate-limit errors: Add delays between requests (limit to 1 request per second)
3. For important tasks: Implement retry logic with exponential backoff

Claude Code-Specific Stalls: Auto Mode, Editing Tools, and MCP

Unlike the web version, Claude Code can stall because of how the agent itself behaves. Once you can name the symptom, most of these clear up in a couple of minutes.

Auto Mode keeps waiting for approval or loops

While working in Auto Mode, you may see the same confirmation prompt over and over, or the session hangs on "stopping your response." Usually the safety classifier has flagged a command as needing confirmation and is waiting for your approval.

When things stall, first check the log to see which tool call is waiting. If it's a command that's safe to allow (running tests, building, and so on), add it to your allowlist with /permissions so the same kind of confirmation no longer blocks you.

If you're stuck in an approval loop, press Esc to interrupt, then give a fresh instruction that names the scope explicitly — for example, "you may automatically run only the tests in this directory." Granting broad permissions increases risk, so it's safer to approve specific actions step by step.

The Edit / Write tools fail and progress stops

A file edit can fail because the target string "isn't found" or "matches more than once," and the session stops there. Most often the file's latest contents weren't read before the edit, or the string you're replacing isn't unique.

The fix is simple: re-read the file before editing, and make your target unique by including enough surrounding lines. An edit can fail just because indentation or a space doesn't match the real file, so watch for subtle differences introduced when copying text.

If it stops on a permission error (can't write), check whether the file is read-only or locked by another process.

Can't connect to an MCP server, or it times out

When an MCP server won't start, its tools don't appear in the list, or the connection times out, it's almost always a configuration or startup-order issue.

Start with claude mcp list to check each server's connection state. If one shows failed, review whether the command path and arguments in your config match your actual environment. Pointing at a relative path or an uninstalled executable fails the connection on the spot.

If the server starts but its tools aren't usable, restart Claude Code so it reloads the MCP servers. When a server is slow to initialize, setting a longer timeout for the first connection helps. For servers with authentication (such as OAuth), an expired token can cause a sudden disconnect, so re-authenticating is worth adding to your checklist.

Browser and App-Specific Fixes

Claude on Chrome Not Working?

Chrome is Claude's most optimized environment, but extensions can interfere.

Solution:

• Go to Settings → Extensions and disable all extensions temporarily
• Access Claude and confirm it works
• Re-enable extensions one by one to identify the culprit

Claude on Safari Issues?

Safari may have compatibility issues with some of Claude's JavaScript features.

Solution:

• Go to Settings → Privacy and enable "Prevent cross-site tracking"
• Clear Safari's cache and stored data
• Consider switching to Chrome for more reliable performance

Claude Mobile App Won't Respond?

The official Claude mobile apps occasionally have bugs that are fixed in updates.

Solution:

• Update to the latest version in the App Store or Google Play
• Delete and reinstall the app from scratch
• Use the web version (claude.ai) as a temporary workaround

Catching a Stall Inside Unattended Automation — Telling a Hang from Slowness

When you're chatting with Claude in real time, a slow response just means waiting a little longer. Wire Claude into unattended automation, though, and the calculus changes. A single call that never returns can quietly freeze an entire pipeline while no one is watching.

As an indie developer, I run scheduled jobs for my own Dolice blogs and app backends. One morning a single API call hung, and the whole batch that was supposed to finish before sunrise had simply stopped. No error — just a "started" line in the log and nothing after it. That silence turned out to be the scariest failure mode of all.

The starting point is to separate three states:

• Slow — it takes a while but eventually returns. Usually a long input or a busy server.
• Hang — it never returns. The connection is held open, waiting forever.
• Rate limited — a 429 comes back and you're explicitly refused. Waiting fixes it.

In a chat you can feel the difference between slow and hung. In automation, your code has to detect it — and the key is to stop waiting. Always set a hard timeout, and when it's exceeded, stop, log it, and retry a bounded number of times. That one change turns a hang from a "silent stop" into a visible, retryable event.

import time
from anthropic import Anthropic
 
client = Anthropic(api_key="YOUR_API_KEY")
 
def ask_with_timeout(prompt, hard_timeout=60, max_retries=2):
    """A minimal pattern for catching hangs and retrying.
    Never wait forever: abort past the limit and record it."""
    last_error = None
    for attempt in range(max_retries + 1):
        try:
            # Passing timeout keeps you from holding a dead connection open
            resp = client.messages.create(
                model="claude-sonnet-4-6",
                max_tokens=1024,
                messages=[{"role": "user", "content": prompt}],
                timeout=hard_timeout,
            )
            return resp.content[0].text
        except Exception as e:
            last_error = e
            # Exponential backoff: rate limits and transient errors often clear here
            wait = 2 ** attempt
            print(f"attempt {attempt + 1} failed: {type(e).__name__} — waiting {wait}s")
            time.sleep(wait)
    # If every attempt fails, surface it instead of swallowing it
    raise RuntimeError(f"timed out after {max_retries + 1} attempts") from last_error

The crucial part is always specifying timeout. Omit it and your process will wait on a dead connection indefinitely, freezing the unattended run right there. With a ceiling in place, a hang surfaces as an exception you can put on a backoff-and-retry path.

Raising instead of swallowing the final failure is deliberate too. In automation, quietly eating an error is the hardest kind of bug to find. If it stalled, say so. Since I started treating it that way, my morning jobs stopped dying in silence.

Looking back

Most Claude issues—whether it's not responding, running slowly, or stopping mid-response—have straightforward solutions. The key is systematically diagnosing the symptom, testing your hypothesis, and applying the right fix.

If you've worked through this guide and still have issues, reach out to Anthropic's support team with specific error messages and details about when the problem occurs. They're responsive and genuinely helpful with edge cases.