Unlimited Cursor with the Claude API

Cursor is one of the most popular AI-powered code editors, and it supports bring-your-own-key through an OpenAI base URL override. By pointing that override at a flat-rate unlimited Claude gateway, you get full Claude access inside Cursor without per-token billing eating into your workflow. This guide covers the complete setup, advanced configuration, model selection strategy, troubleshooting common failures, and a concrete explanation of why unlimited access transforms the Cursor experience for heavy users.

Prerequisites and What You Need

Before configuring Cursor to use an unlimited Claude gateway, you need two things: a working Cursor installation (version 0.40 or later supports the current settings layout) and an active API key from AI Prime Tech Unlimited. The gateway exposes an OpenAI Chat Completions-compatible endpoint, which is exactly what Cursor's custom-key infrastructure expects. This means you connect Claude through Cursor's OpenAI fields, not the separate Anthropic section.

Have your base URL ready: https://unlimited.aiprimetech.io/v1 — note the /v1 suffix, because Cursor appends /chat/completions to whatever you enter. Your API key is the one from your AI Prime Tech Unlimited dashboard. Cursor performs a live verification call when you click Verify, so the gateway must be reachable at that moment. If you are behind a corporate proxy or VPN, confirm that outbound HTTPS to the gateway domain is allowed.

One important architectural detail: Cursor's custom-key path is global. Once you enable the override, all custom-model chat traffic routes through your base URL. If you also use other providers in Cursor, be aware that the override applies across the board for any model that goes through the OpenAI path.

Step-by-Step Base URL and Key Configuration

Open Cursor Settings using the gear icon in the sidebar or the keyboard shortcut Ctrl+Comma (Cmd+Comma on macOS). Click Models in the left navigation panel. Scroll to the OpenAI API Key section. You will see two fields: one for the base URL override and one for the API key itself.

Enable the Override OpenAI Base URL toggle and paste https://unlimited.aiprimetech.io/v1 into the field. In the OpenAI API Key field directly below, paste your AI Prime Tech Unlimited key. The field label says OpenAI, but the request is routed to your custom endpoint, which forwards it to Claude. This is how Cursor handles all custom providers — through the OpenAI-compatible path.

Do not add /chat/completions to the base URL. Cursor appends that path segment automatically. If you accidentally include it, you will get a 404 error that looks like an authentication problem but is actually a path-doubling issue. The correct value is always the /v1 root and nothing more.

Adding Claude Models and Verification

With the base URL and key in place, click the + Add Model button. Type the exact model identifier that the gateway exposes — for example claude-sonnet-4-5, claude-opus-4-6, or claude-haiku-4-5. The string must match what the gateway expects in the request's model field verbatim. Cursor does not translate or alias model names for you.

Make sure the model toggle is enabled (green). Then click Verify. Cursor sends a real API call to your base URL with your key and the model you just added. If the URL path is correct, the key is valid, and the gateway is up, the verification succeeds and you can click Save. If it fails, check the three most common causes: missing /v1 suffix, wrong model string, or expired key.

After saving, open a new chat with Ctrl+L (Cmd+L on macOS), select your Claude model from the model picker dropdown at the top of the chat panel, and send a test message. If you get a response, the end-to-end connection is confirmed and you are running Claude through your unlimited gateway.

Advanced Workflows and Model Selection Strategy

For daily coding work in Cursor — writing functions, debugging, explaining code, generating tests — Claude Sonnet is the practical default. It balances speed and reasoning quality well enough for the vast majority of editing tasks. Reserve Claude Opus for moments when you need deeper architectural reasoning, complex refactoring plans, or analysis of large interconnected codebases where Sonnet might miss subtle dependencies.

You can add multiple Claude models to Cursor simultaneously. Keep claude-sonnet-4-5 as your everyday model and add claude-opus-4-6 for when you need heavier reasoning. Switch between them in the model picker per-conversation. On an unlimited plan there is no cost penalty for using the stronger model when the task demands it — the only tradeoff is slightly higher latency for Opus responses.

For rapid iteration cycles — quick questions, short completions, formatting help — Claude Haiku provides the fastest response times. Add claude-haiku-4-5 as a third option for those lightweight interactions where speed matters more than depth. The unlimited plan covers all three tiers equally, so you can freely pick the right tool for each moment.

Known Limitations with Custom Keys in Cursor

Several Cursor features behave differently on any custom key, regardless of provider. Cursor Tab (the inline autocomplete that appears as you type) always uses Cursor's own built-in model and is never routed to your gateway. This is a Cursor-side restriction, not a gateway limitation. Your unlimited key powers the chat, edit, and manual completion features only.

Agent mode and Composer mode have restrictions with custom keys. The Ask, Plan, and Chat modes work reliably through the custom key path. Agentic editing — where Cursor autonomously makes changes across files — depends on full tool-call support and is limited on Cursor's side regardless of what provider you use. If you need deep agentic editing, consider pairing Cursor with a dedicated agent like Claude Code or Cline for those tasks.

There is no request-timeout setting exposed in Cursor's UI. If you select a model that is slow to respond (Opus on a complex query, for example), the chat may appear to hang. This is normal — wait for the response or switch to a faster model for tight feedback loops. On the unlimited plan you are not penalized for these longer responses since there is no per-token cost accumulating.

Troubleshooting Common Issues

The most frequent failure is a 404 error during verification. This almost always means the base URL path is wrong. Confirm you have exactly https://unlimited.aiprimetech.io/v1 with no trailing slash and no additional path segments. Cursor appends /chat/completions itself, so the final URL becomes https://unlimited.aiprimetech.io/v1/chat/completions — if you included /chat/completions in the field, the actual request hits a doubled path that does not exist.

If verification succeeds but chat responses fail, check that you selected the correct model in the chat picker. Cursor can silently fall back to a built-in model if you forget to explicitly pick your custom model after enabling the override. Always confirm the model name shown in the picker matches what you configured.

Authentication errors (401 or 403) mean the key is wrong, expired, or does not have access to the requested model. Copy the key fresh from your dashboard, paste it without leading or trailing spaces, and retry. If the gateway is temporarily down for maintenance, you will see connection errors — these resolve once the service is back.

Why Unlimited Access Transforms Cursor Usage

A productive day in Cursor easily generates hundreds of API requests. Every chat message, every inline edit, every code explanation, every test generation — each is a separate API call with input tokens (your code context) and output tokens (the model's response). On per-token billing, a heavy Cursor day can cost more than many developers expect, especially when working with large files or long conversations.

The unlimited model eliminates this friction entirely. You set up the gateway once and then use Cursor as intensively as your work demands. There is no meter to watch, no mental calculation about whether a question is worth asking, no hesitation before sending a large file for review. The flat-rate subscription covers all of it for one predictable monthly price.

This is particularly valuable during intense development phases — shipping a feature, debugging a production issue, onboarding onto an unfamiliar codebase. These are exactly the moments when you want to lean heavily on Claude without worrying about cost. Unlimited access means the tool is always there at full capacity when you need it most.

# Cursor Settings -> Models -> OpenAI API Key
#
# 1. Enable: Override OpenAI Base URL
# 2. Base URL: https://aiprimetech.io/v1
# 3. API Key:  <your AI Prime Tech Unlimited key>
# 4. + Add Model: claude-sonnet-4-5
#    (optionally add claude-opus-4-6 and claude-haiku-4-5)
# 5. Click Verify -> Save
# 6. New Chat (Ctrl+L) -> pick your Claude model -> start coding
#
# The gateway speaks OpenAI Chat Completions format.
# Cursor appends /chat/completions to the base URL automatically.

Frequently asked questions

Do I use the Anthropic field or the OpenAI field in Cursor?
Use the OpenAI field with Override OpenAI Base URL enabled. The AI Prime Tech Unlimited gateway is OpenAI-compatible, so Cursor's OpenAI custom-key path is correct. The separate Anthropic field in Cursor is only for direct api.anthropic.com connections and does not support custom base URLs the same way.

Why did Verify fail with a 404 error?
The base URL is wrong. Enter exactly https://unlimited.aiprimetech.io/v1 — no trailing slash, no /chat/completions suffix. Cursor appends /chat/completions itself. A doubled path or missing /v1 causes the 404. Also confirm the model string matches what the gateway exposes.

Will Cursor Tab autocomplete work with my unlimited key?
No. Cursor Tab always uses Cursor's own built-in model regardless of your custom key configuration. Your unlimited key powers chat, edit, and manual completion features. Tab autocomplete is a separate Cursor-internal feature.

Can I use multiple Claude models in Cursor simultaneously?
Yes. Add multiple models (Sonnet, Opus, Haiku) and switch between them in the chat picker per conversation. On the unlimited plan there is no cost difference between models — pick based on the task's speed and reasoning requirements.

Is there a per-request cost when using the unlimited plan in Cursor?
No. The flat-rate subscription covers all eligible usage during your plan period. There are no per-token or per-request charges — only fair-use rate limits to keep the gateway responsive for all users.

AI Prime Tech is an independent API gateway. It is not affiliated with, endorsed by, or a reseller of Anthropic. Claude and related model names are trademarks of their respective owners.