Real-time context monitoring.
Inside your editor.

Track token budgets with real tokenizers, score tab relevance, and manage context — available as a VS Code extension, MCP server, and CLI. v3.1.3

Install from Marketplace View Source on GitHub

ext install vfaraji89.tokalator

12Features

5Settings

VS Code 1.99+Minimum version

MITLicense

156Total acquisitions

82.98%Conversion rate

220+Community devs

124Unit tests

First 30 days (Feb 4 – Mar 6, 2026) · 82.98% page-to-acquisition conversion · 124 tests across 6 files

~50iLab / Kariyer.net— Internal deployment — engineers used the extension for one week

~90SaaS Bridge— Istanbul, March 4 2026

~80TechCareer— Istanbul, March 5 2026

Screenshots

Tokalator sidebar dashboard showing token budget, next turn preview, and budget breakdown — Sidebar dashboard — live token budget, next turn preview, and budget breakdown

Tokalator /breakdown command in chat showing file token counts and relevance scores — @tokalator /breakdown — per-file token counts and relevance scores in chat

Tokalator chat command autocomplete showing all @tokalator slash commands — 11 chat commands via @tokalator — full context management workflow

Features

▣ Token Budget Dashboard

Activity bar sidebar panel showing real-time token usage, budget meter, per-file breakdown, and active tokenizer info.

View via sidebar iconVS Code

◎ @tokalator Chat Participant

Eleven chat commands for inline budget management directly in Copilot Chat.

@tokalator /count, /optimize, /breakdown, /pin, /unpin, /instructions, /model, /compaction, /preview, /reset, /exitVS Code

⬡ Tab Relevance Scoring

Ranks open tabs by relevance using import analysis, path similarity, edit recency, and diagnostics.

tokalator.optimizeVS Code

⊕ Model-Specific Tokenizers

Real BPE tokenizers per provider — Claude BPE for Anthropic, o200k_base for OpenAI. No more guessing at 4 chars per token.

@tokalator /modelVS Code

◇ 17 Model Profiles

Built-in profiles for Claude Opus/Sonnet/Haiku 4.x, GPT-5.4, GPT-5.4 Mini, GPT-5.2/5.1 Codex, GPT-4.1, o3, o4-mini, Gemini 3.x Pro/Flash, Gemini 2.5 Pro. Context windows and rot thresholds auto-configured.

Settings: tokalator.modelVS Code

⚠ Context Rot Warnings

Warns when conversation turns exceed the model's threshold. Each model has its own rot limit based on benchmarks.

AutomaticVS Code

⊕ Pinned Files

Pin important files so they are always treated as high-relevance in context scoring.

@tokalator /pin <file>VS Code

◈ Instruction File Scanner

Detects and tokenizes instruction files silently injected into every prompt: .github/copilot-instructions.md, CLAUDE.md, AGENTS.md, .cursorrules, and .instructions.md. Reveals hidden per-file token cost.

@tokalator /instructionsVS Code

▹ Next Turn Preview

See estimated token cost of your next message before sending. Dashboard shows a live preview box with growth projections and overflow warnings.

@tokalator /previewVS Code

◉ Session Management

Track per-turn context growth with compaction analysis. Reset sessions, save summaries, and see last session stats on activation.

@tokalator /compaction, /reset, /exitVS Code

⬡ MCP Server for Claude Code

Brings real Claude BPE token counting into Claude Code and any MCP-capable agent via stdio transport. Four tools: count_tokens, estimate_budget, preview_turn, list_models. Zero API calls — fully offline.

claude mcp add --transport stdio tokalator -- node tokalator-mcp/build/index.jsMCP

◆ CLI — tokalator count, budget, preview, models

Standalone terminal tool for non-VS Code workflows. Count tokens in files, get full budget breakdowns, preview next-turn cost, and list Claude model profiles — all from the command line.

npm install -g . (from tokalator-mcp/) · tokalator count src/main.ts · tokalator budget src/**/*.tsCLI

Configuration

Setting	Default	Description
`tokalator.model`	claude-opus-4.6	AI model to calculate context budget against. Sets tokenizer, context window, and rot threshold automatically.
`tokalator.relevanceThreshold`	0.3	Tabs below this relevance score are marked as distractors (0–1)
`tokalator.windowSize`	1,000,000	Override context window size in tokens. Leave at default to use the selected model's window.
`tokalator.contextRotWarningTurns`	20	Warn about context rot after this many chat turns
`tokalator.autoRefreshInterval`	2000	Dashboard refresh interval in milliseconds

MCP Server & CLI

Real Claude BPE token counting inside Claude Code via stdio MCP transport, plus a standalone terminal CLI for SSH sessions, containers, and non-VS Code workflows.

tokalator — cli

○user@machine ~/my-repo $ tokalator

Welcome to Tokalator

█████  ███  █   █   █   █       █   █████  ███  ████ 
  █   █   █ █  █   █ █  █     █ █    █   █   █ █   █
  █   █   █ ███   █████ █     █████   █   █   █ ████ 
  █   █   █ █  █  █   █ █     █   █   █   █   █ █ █  
  █    ███  █   █ █   █ █████ █   █   █    ███  █  █

CLI Version

~/my-repoclaude-opus-4.6 (3×)

>Type @ to mention files, / for commands, or ? for shortcuts

shift+tab switch mode

Architecture

6 layers inside the extension:

Core Engine

Subscribes to VS Code editor events (active editor, tab open/close, document edits, diagnostics). Builds ContextSnapshot records, manages pinned files and model selection via workspaceState. All handlers debounced 300 ms.

Tokenizer Service

Provider-specific BPE tokenizers — claude-tokenizer for Anthropic, o200k_base (js-tiktoken) for OpenAI, heuristic (~4 chars/token) for Gemini. Lazy-loaded and cached per document URI + version.

Relevance Scorer

Scores each open tab R ∈ [0, 1]: language match (0.25), import analysis (0.30), path similarity (0.20), edit recency (0.15), diagnostics (0.10). Pinned and active tabs override to R = 1.0.

Context Optimizer

Identifies distractor tabs (R < 0.3) and closes them via /optimize or the Optimize Tabs button. Reduces attention dilution by removing files unlikely to contribute to the current task.

Chat Participant

@tokalator with 11 slash commands — /count, /optimize, /breakdown, /pin, /unpin, /instructions, /model, /compaction, /preview, /reset, /exit.

MCP Server & CLI (tokalator-mcp)

Separate component: exposes count_tokens, estimate_budget, preview_turn, list_models to Claude Code via stdio MCP transport. Also ships a standalone CLI binary for terminal workflows.

Tokenizers

Each provider uses its own tokenizer for accurate token counts:

Provider	Models	Tokenizer	Source
Anthropic	Opus 4.6, Opus 4.5, Sonnet 4.6, Sonnet 4.5, Sonnet 4, Haiku 4.5	`Claude BPE`	@anthropic-ai/tokenizer
OpenAI	GPT-5.4, GPT-5.4 Mini, GPT-5.2 Codex, GPT-5.1 Codex, GPT-4.1, o3, o4-mini	`o200k_base`	js-tiktoken
Google	Gemini 3.1 Pro, Gemini 3 Pro, Gemini 3 Flash, Gemini 2.5 Pro	`Heuristic`	~4 chars/token (no official JS tokenizer)

Installation

◉ Marketplace

ext install vfaraji89.tokalator

Requires VS Code 1.99+

◇ VSIX

Download tokalator-3.1.3.vsix from releases
Cmd+Shift+P → Extensions: Install from VSIX...
Select the downloaded .vsix file

code --install-extension tokalator-3.1.3.vsix

◆ Build from Source

git clone https://github.com/vfaraji89/tokalator.git
cd tokalator/tokalator-extension-vs
npm install
npm run compile

◆ CLI (npm)

Requires Node.js 18+

git clone https://github.com/vfaraji89/tokalator.git
cd tokalator/tokalator-mcp && npm install && npm run build
npm install -g .

tokalator count src/main.ts
tokalator budget src/**/*.ts
tokalator preview 42000
tokalator models

⬡ MCP Server (Claude Code)

Requires Node.js 18+ and Claude Code

Build the CLI first (see above)
claude mcp add --transport stdio tokalator -- node tokalator-mcp/build/index.js
In Claude Code: /mcp → verify tokalator is listed

claude mcp add --transport stdio tokalator -- node tokalator-mcp/build/index.js