Unlimited Cline with the Claude API

Cline is a fully autonomous coding agent that runs inside VS Code, capable of reading files, writing code, executing terminal commands, and browsing the web — all through Claude. Each task involves multiple round-trips as Cline plans, acts, and verifies its work, making it one of the heaviest consumers of API tokens among developer tools. Connecting Cline to a flat-rate unlimited Claude gateway eliminates the cost anxiety that comes with this intensity and lets you use Cline at its full potential without watching a billing meter.

Why Cline Burns Through Tokens So Fast

Cline operates in a Plan-then-Act loop. When you give it a task, it first reasons about what needs to happen (Plan mode), then executes each step individually (Act mode). Every file read, every terminal command, every code edit is a separate API call that includes the full conversation history up to that point. A single task that touches five files might generate twenty or more API calls, each carrying an increasingly large context.

On top of this, Cline supports MCP (Model Context Protocol) servers that inject additional tools — database queries, documentation lookups, web searches — each adding tool-call overhead to every request. When you enable Computer Use for browser automation, each screenshot and interaction is another round-trip. The token consumption compounds quickly, and a complex refactoring session can easily hit six figures in total tokens within minutes.

This is exactly why an unlimited plan makes sense for serious Cline users. The tool is designed to be autonomous and thorough, which means it should not be throttled by cost concerns. Flat-rate access lets Cline do what it does best — iterate aggressively until the task is complete.

Connecting via the Anthropic Provider Path

Open the Cline extension settings in VS Code (click the gear icon in the Cline sidebar panel). In the API Provider dropdown, select Anthropic. You will see fields for the API Key and an optional Base URL. Enter your AI Prime Tech Unlimited key in the API Key field. For the Base URL, enter https://unlimited.aiprimetech.io — note that this is the root host without /v1, because the Anthropic SDK appends /v1/messages automatically.

Select your model from the Model dropdown. If you do not see the model you want, type it directly — Cline allows free-text model IDs. For most work, claude-sonnet-4-5 is the right choice. For complex architectural tasks, switch to claude-opus-4-6. After entering the details, Cline will make a test request to verify the connection.

The Anthropic path is the cleanest integration because Cline's tool-calling implementation is optimized for the Anthropic Messages API format. Tool results, thinking blocks, and streaming all work natively without translation layers. If you have no specific reason to use the OpenAI path, prefer this one.

Connecting via the OpenAI-Compatible Provider Path

If you prefer the OpenAI-compatible route, select OpenAI Compatible from the API Provider dropdown. Set the Base URL to https://unlimited.aiprimetech.io/v1 — here the /v1 suffix IS required because the OpenAI-compatible path does not auto-append it. Enter your API key in the key field and type the model identifier (e.g., claude-sonnet-4-5) in the Model ID field.

With the OpenAI-compatible path, you must manually configure two additional fields that the Anthropic path handles automatically: Context Window and Max Output Tokens. Set Context Window to 200000 (200k) for Sonnet/Opus or 200000 for Haiku. Set Max Output Tokens to 8192 for standard responses or 16384 if you want longer outputs. These values control how Cline manages conversation truncation and output limits.

The OpenAI-compatible path works well but lacks native support for some Anthropic-specific features like extended thinking. For most Cline workflows — file editing, terminal commands, code generation — both paths perform identically. Choose based on whether you need Anthropic-native features or prefer the OpenAI format for compatibility with other tools sharing the same configuration.

Configuring Plan Mode and Checkpoints

Cline's Plan/Act toggle at the top of the chat panel controls whether the agent plans before acting or jumps straight into execution. With unlimited access, you should generally leave Plan mode enabled. It adds one extra API call per task to generate the plan, but the improved accuracy and reduced wasted actions more than compensate. Without unlimited access, users often disable planning to save tokens — that tradeoff disappears here.

Checkpoints are Cline's safety net: it creates shadow git commits at key points so you can roll back if the agent makes a mistake. This feature works independently of the API provider and requires no gateway configuration. Enable it in Cline settings under Checkpoints. With unlimited access, you can let Cline attempt more ambitious changes knowing that rollback is one click away and retrying costs nothing extra.

For complex tasks, consider using the Ask mode first — type your task and let Cline ask clarifying questions before it starts working. This front-loads the conversation context and produces better plans. On unlimited, the extra back-and-forth costs nothing and significantly improves outcomes for ambiguous requirements.

MCP Servers and Tool Expansion

Cline supports MCP servers that extend its capabilities beyond file and terminal operations. You can connect database servers, documentation indexes, web scrapers, and custom business-logic tools. Each MCP tool appears as an available action in Cline's tool-calling repertoire, and the agent will use them when relevant to the task.

To configure MCP servers, open Cline settings and navigate to the MCP Servers section. Add server entries with their command, arguments, and environment variables. Cline automatically discovers the tools each server exposes and includes them in the system prompt sent with every request. This means more tools equals more tokens per request — another reason unlimited access is valuable for MCP-heavy workflows.

Popular MCP servers for Cline include filesystem watchers (for monitoring build output), database clients (for querying schemas during ORM work), and documentation servers (for pulling API references into context). The combination of autonomous agent behavior and rich tool access makes Cline extremely capable but also extremely token-hungry.

Computer Use and Browser Automation

Cline supports Anthropic's Computer Use capability for browser automation tasks. Enable it in the Cline settings under the Computer Use toggle. When active, Cline can take screenshots of browser windows, click elements, type text, and navigate pages — useful for testing web applications, filling forms, or scraping data that requires JavaScript rendering.

Each Computer Use interaction is a multi-step process: Cline takes a screenshot (sent as a base64 image in the API request), reasons about what it sees, decides on an action (click, type, scroll), executes it, takes another screenshot, and repeats. A single browser task can easily generate dozens of API calls, each carrying image data that inflates token counts significantly. This is one of the most token-intensive features in any AI coding tool.

On a per-token plan, Computer Use sessions can become expensive fast. On unlimited, you can let Cline drive the browser as much as the task requires without concern. This makes it practical for end-to-end testing workflows, visual regression checks, and complex web scraping tasks that would be prohibitively expensive on metered billing.

Troubleshooting Cline Connection Issues

If Cline reports a connection error immediately after configuration, check the base URL format first. For the Anthropic provider: use the root host (https://unlimited.aiprimetech.io) without /v1. For OpenAI Compatible: use https://unlimited.aiprimetech.io/v1 with the suffix. Mixing these up is the most common configuration mistake.

If requests work initially but fail mid-task, you may be hitting fair-use rate limits. The unlimited plan has rate limits to keep the gateway responsive for all users — these are generous but not infinite. If Cline is making rapid-fire requests during a large refactoring task, it may temporarily exceed the per-minute limit. The gateway returns a 429 status with a Retry-After header; Cline handles this gracefully by waiting and retrying.

For model-not-found errors, verify that the model string you entered matches exactly what the gateway supports. Model identifiers are case-sensitive and version-specific. Check the AI Prime Tech documentation for the current list of supported model IDs.

# Cline Settings -> API Provider: Anthropic
#
# Base URL: https://aiprimetech.io
# API Key:  <your AI Prime Tech Unlimited key>
# Model:    claude-sonnet-4-5
#
# --- OR ---
#
# Cline Settings -> API Provider: OpenAI Compatible
#
# Base URL:         https://aiprimetech.io/v1
# API Key:          <your AI Prime Tech Unlimited key>
# Model ID:         claude-sonnet-4-5
# Context Window:   200000
# Max Output:       8192

Frequently asked questions

Should I use the Anthropic or OpenAI Compatible provider in Cline?
Use the Anthropic provider for the cleanest integration. It supports native tool calling, streaming, and thinking blocks without translation. The OpenAI Compatible path works too but requires manual Context Window and Max Output configuration and lacks some Anthropic-specific features.

Why does Cline use so many tokens per task?
Cline's Plan-then-Act architecture means every step — file read, edit, terminal command, verification — is a separate API call carrying the full conversation history. A five-file refactoring task can generate 20+ calls, each with growing context. MCP tools and Computer Use add further overhead.

Do I need to set Context Window manually?
Only on the OpenAI Compatible path. The Anthropic provider path sets context window limits automatically based on the model. For OpenAI Compatible, set it to 200000 for Sonnet and Opus.

Can I use Computer Use with the unlimited plan?
Yes. Enable the Computer Use toggle in Cline settings. Each browser interaction sends screenshots as base64 images, which inflates token usage significantly. The unlimited plan covers this without extra charges — only fair-use rate limits apply.

What happens when I hit rate limits during a long Cline task?
The gateway returns HTTP 429 with a Retry-After header. Cline automatically waits the specified duration and retries the request. You may see a brief pause in the agent's progress, but it resumes without intervention. The limits are per-minute and reset quickly.

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.