[BUG] I am instantly hitting an "API Error: Rate Limit Reached" when I start a new conversation with Claude Code. This occurred after enabling "Fast Mode" with Opus.

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

I am completely locked out of using Claude Code because of an instant 429 Rate Limit error that happens the second I try to start a new conversation.
Specifically, here is what is happening:
The Trigger: Last night I enabled "Fast Mode" with Opus.
The Result: This morning, Claude Code is entirely bricked. Even if I switch to Sonnet and try to start a brand-new, empty conversation, I instantly get a Rate Limit error.
The Root Cause (from the logs): The logs show a 429 error with the message: "Extra usage is required for long context requests."

What Should Happen?

Clean Initialization: When I start a new conversation, Claude Code should start with a fresh, empty context window. It should not carry over massive hidden context or cache from previous sessions (like my Opus "Fast Mode" session from last night).
No Instant 429 Errors: With a Max Subscription and an empty new chat, the application should successfully initialize and wait for my prompt, rather than instantly throwing an "Extra usage is required for long context requests" rate limit error.
Graceful MCP Degradation: While my logs show my GitHub MCP lacks a token, this should just cleanly disable the GitHub tool. It should not contribute to an app-breaking 429 API rate limit error or stop the core CLI from functioning.

Error Messages/Logs

{
    "error": "Error: Plugin MCP server error - mcp-config-invalid: MCP server github invalid: Missing environment variables: GITHUB_PERSONAL_ACCESS_TOKEN\n    at hjH (/$bunfs/root/src/entrypoints/cli.js:2283:23982)\n    at processTicksAndRejections (native:7:39)",
    "timestamp": "2026-03-05T13:46:38.699Z"
  },
  {
    "error": "Error: Plugin MCP server error - mcp-config-invalid: MCP server github invalid: Missing environment variables: GITHUB_PERSONAL_ACCESS_TOKEN\n    at hjH (/$bunfs/root/src/entrypoints/cli.js:2283:23981)\n    at async X (/$bunfs/root/src/entrypoints/cli.js:4166:16475)\n    at processTicksAndRejections (native:7:39)",
    "timestamp": "2026-03-05T13:46:39.081Z"
  },
  {
    "error": "Error: Plugin MCP server error - mcp-config-invalid: MCP server github invalid: Missing environment variables: GITHUB_PERSONAL_ACCESS_TOKEN\n    at hjH (/$bunfs/root/src/entrypoints/cli.js:2283:23981)\n    at async J (/$bunfs/root/src/entrypoints/cli.js:4166:17191)\n    at processTicksAndRejections (native:7:39)",
    "timestamp": "2026-03-05T13:46:39.081Z"
  },
  {
    "error": "Error: Plugin MCP server error - mcp-config-invalid: MCP server github invalid: Missing environment variables: GITHUB_PERSONAL_ACCESS_TOKEN\n    at hjH (/$bunfs/root/src/entrypoints/cli.js:2283:23981)\n    at async X (/$bunfs/root/src/entrypoints/cli.js:4166:16475)\n    at processTicksAndRejections (native:7:39)",
    "timestamp": "2026-03-05T13:46:39.135Z"
  },
  {
    "error": "Error: Plugin MCP server error - mcp-config-invalid: MCP server github invalid: Missing environment variables: GITHUB_PERSONAL_ACCESS_TOKEN\n    at hjH (/$bunfs/root/src/entrypoints/cli.js:2283:23981)\n    at async J (/$bunfs/root/src/entrypoints/cli.js:4166:17191)\n    at processTicksAndRejections (native:7:39)",
    "timestamp": "2026-03-05T13:46:39.139Z"
  },
  {
    "error": "Error: 429 {\"type\":\"error\",\"error\":{\"type\":\"rate_limit_error\",\"message\":\"Extra usage is required for long context requests.\"},\"request_id\":\"req_011CYk34CBGDDJbfA3k3uxHy\"}\n    at generate (/$bunfs/root/src/entrypoints/cli.js:292:9185)\n    at makeRequest (/$bunfs/root/src/entrypoints/cli.js:331:5435)\n    at processTicksAndRejections (native:7:39)",
    "timestamp": "2026-03-05T13:46:41.730Z"
  }
]

Steps to Reproduce

Have a Max Subscription and configure Claude Code with the GitHub MCP server (without the GITHUB_PERSONAL_ACCESS_TOKEN environment variable set).
Start a session and enable "Fast Mode" using the Opus model.
End the session.
Attempt to start a brand-new conversation later (e.g., the next morning).
Switch the model to Sonnet 4.6 to see if it resolves the issue.
Observe that the CLI instantly crashes/fails to initialize with an API Error: Rate Limit Reached message.
Check the logs and observe the 429 error stating "Extra usage is required for long context requests."

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

2.1.69

Claude Code Version

2.1.69

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Terminal.app (macOS)

Additional Information

Request ID for Backend Tracing: The most helpful piece of information for your backend team is likely the request_id from the 429 error log: req_011CYk34CBGDDJbfA3k3uxHy. This should allow you to look up exactly what payload the CLI was trying to send on initialization.
Subscription Level: I am on the Max Subscription. I should not be hitting a hard limit on an empty, brand-new conversation.
State/Cache Suspicion: The bug behaves as if "Fast Mode" combined with the Opus model permanently corrupted or bloated my local session state. Even when I ask the CLI to use Sonnet 4.6 for a completely new conversation this morning, it acts as if it is trying to process a massive context window.
The MCP Error: There is a secondary error regarding the GITHUB_PERSONAL_ACCESS_TOKEN. I suspect this is a red herring and unrelated to the 429 Rate Limit crash, but I am including it just in case the failing MCP server causes an infinite retry loop that spikes API usage in the background.