Frequently asked questions

Frequently asked questions

What is Tokenade?

Tokenade is a local MCP proxy that automatically cuts the token usage of AI agents (Claude Code, Cursor, Codex CLI, GitHub Copilot, Kilo, Windsurf, Cline, and more). It intercepts agent–LLM traffic, filters noisy shell output, replaces whole-file reads with skeleton or semantic search, and loads MCP tools lazily — without touching your code or prompts.

Which agents and IDEs does Tokenade work with?

Tokenade works with any MCP-compatible agent: Claude Code, Cursor, GitHub Copilot (agent mode), Kilo Code, Windsurf (Cascade), Cline, Codex CLI, and any other MCP client. OS support: Linux, macOS, and Windows.

How much does Tokenade cost?

A 7-day free trial with no limits, then 1% (excl. VAT) of the API value of the tokens Tokenade saves you, at the price of the model that did the work — for example $0.02 per million tokens saved on Sonnet. No subscription fee, no cap, unlimited machines. A card is asked at signup; nothing is billed during the trial and you can cancel anytime. Per-model rates are on the pricing page (tokenade.net/en/pricing). For teams, Enterprise adds a Tokenade proxy with governance: book a call from the pricing page.

What happens if my payment fails or I cancel?

Tokenade passes your commands through unchanged (no optimization) on all your machines. Nothing breaks — you only lose the savings. Add or update your card from the dashboard to resume.

How many machines can I use?

There is no limit. `tokenade login` links each machine to your account (required) so its savings show up in your dashboard and on your bill.

Does Tokenade send my code or prompts to external servers?

No. Tokenade runs entirely locally. The only telemetry it sends is token-savings metadata: operation type + before/after token counts. No code, no prompts, no file paths are transmitted. Any secrets that appear in shell output are redacted before logging.

Which operating systems does Tokenade support?

Linux, macOS, and Windows are all officially supported. Installation is via npm/npx — no native binary required.

How do I install and activate Tokenade?

Three commands: `npm install -g @tokenade/cli`, `tokenade install`, then `tokenade login`. The second detects your agents and configures them automatically (hooks + MCP); the third connects your account (required) to activate the optimizations and credit your savings to the dashboard. No account yet? Create one: 7-day free trial, card required, nothing billed during the trial.

How many tokens does Tokenade actually save?

Savings depend on the agent and project. On typical Claude Code development sessions, users report 40–80% fewer tokens consumed. The Tokenade dashboard shows your real-time savings.

Is Tokenade open source?

No. Tokenade is a proprietary product: the binary is free to install, its source is not public. What you can verify yourself is the part that matters: compaction runs entirely on your own machine, and the only thing that reaches our servers is usage accounting — how many tokens were saved, on which model. That is observable at the network level, so you never have to take our word for it.

How is Tokenade different from other token optimizers?

Most optimizers work at the prompt level (truncation, summarization). Tokenade works at the MCP transport layer: it structurally replaces expensive tools (full file reads, verbose bash) with compact equivalents (skeleton, semantic_search, exec_compact). The result is systematic reduction without quality loss — not an approximate summary.

How do I cancel?

From the Tokenade dashboard, “Manage billing” (Stripe portal), or email [email protected]. Cancellation takes effect at the end of the current period; that period's usage is billed at 1% of what was saved.