Wick MCP
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Wick MCP show my current session stats"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Wick MCP — Claude token tracker for Desktop & Code
A local MCP server that meters Claude Desktop, Claude Code, and the CLI the same way the Wick Chrome extension meters claude.ai. It accumulates tokens, computes cost (USD + INR) and burn rate, estimates messages left, and streams it all to a live dashboard at http://localhost:6789.
Everything runs locally. No API keys, no accounts, no data leaves your machine.
How Wick MCP works
Wick tracks usage in two ways, so you pick effort vs. coverage:
Client | How it tracks | Effort | Accuracy |
Claude Code / CLI | Automatic. Wick reads the usage logs Claude Code already writes ( | Zero — install once, nothing else | Exact (real API numbers, incl. cache) |
Claude Desktop | Claude calls the | Paste a one-time custom-instruction | Approximate (model self-reports tokens) |
Either way the numbers land in the same session, the same dashboard, and the
same wick_status / wick_summary tools. The server auto-starts the Claude
Code watcher on boot — so for coding sessions, it just works.
Related MCP server: nikhilnt
Install in 30 seconds
One command — installs from npm and patches your Claude Desktop config automatically:
npx usewick-mcp installThen restart Claude Desktop.
git clone https://github.com/pranshuguptaa/wick-mcp
cd wick-mcp
npm install
npm run install-desktopVerify it's working
Open Claude Desktop and type:
"Call wick_status"
You should see your session stats in a boxed panel.
Add to Claude Code (one command)
claude mcp add wick node /absolute/path/to/wick-mcp/server.jsThat's the whole setup. Because Wick auto-reads Claude Code's logs, you do
not need any custom instructions — start (or continue) any coding session and
the dashboard fills in with real usage. Prefer a config file? Add the same block
to .claude/settings.json:
{
"mcpServers": {
"wick": { "command": "node", "args": ["/absolute/path/to/wick-mcp/server.js"] }
}
}Make Claude Desktop track automatically
Claude Desktop has no readable usage log, so it reports via the wick_track
tool. Paste this once into Settings → Custom Instructions:
After every response, silently call wick_track with:
model: the model you used
input_tokens: your input token count
output_tokens: your output token count
cache_read_tokens: cache read tokens if any
Do this without mentioning it.
(Claude Code users can skip this entirely — the log watcher covers it.)
Feed data manually (demos / other clients)
Any tool can push a turn over HTTP — handy for screenshots or wiring up a client Wick doesn't natively watch:
curl -X POST http://localhost:6789/api/track -H 'content-type: application/json' \
-d '{"model":"claude-sonnet-4-6","input_tokens":1200,"output_tokens":400,"cache_read_tokens":0}'Dashboard
Visit http://localhost:6789 while Wick is running (Claude Desktop launches the server automatically once installed). It shows:
Messages left, session cost (₹ + $), and tokens as live stat cards
Two arc gauges — session usage (amber) and daily-budget spend (purple)
A live turn-history table (last 20 turns), updated over WebSocket
A ● LIVE amber dot shows when the dashboard is connected; it auto-reconnects
every 3 seconds if the server restarts.
Tools available
Tool | What it does |
| Track tokens for this response |
| See current session stats |
| Start a fresh session |
| Export data as CSV / JSON / Markdown |
| All-time usage totals |
Ports & storage
6789 — HTTP dashboard +
GET /api/status,GET /api/summary,POST /api/track6790 — WebSocket live updates
~/.wick/history.json— persisted session history (driveswick_summary)CSV exports are written to
~/wick-export-<timestamp>.csvReads (never writes)
~/.claude/projects/**/*.jsonlfor the Code watcher
Environment overrides: WICK_HTTP_PORT, WICK_WS_PORT, WICK_NO_WATCH=1
(disable the Code watcher), WICK_CLAUDE_PROJECTS (custom log path).
How cost is computed
Prices are per million tokens (see models.js). Cache reads are billed at
10% of the input price, and Wick reports the savings versus paying full price.
INR is USD × 84.
Model | Input $/M | Output $/M | Burn |
Opus 4.8 / 4.7 / 4.6 | 15 | 75 | 5.0 |
Sonnet 4.6 | 3 | 15 | 2.0 |
Haiku 4.5 | 0.8 | 4 | 1.0 |
Messages-left estimate: Haiku ≈ 50 messages per 5-hour window, scaled down by the burn multiplier (Sonnet ≈ 25, Opus ≈ 10), minus messages already used.
Troubleshooting
wick_statussays the tool isn't found — fully quit and reopen Claude Desktop (MCP servers only load at startup).Dashboard won't load — the server only runs while a Claude client (or
node server.js) is running. Check that ports 6789/6790 are free.Claude Code usage isn't showing — the watcher tracks turns that happen after Wick starts (it never back-fills history). Send one message and it appears. Confirm
~/.claude/projects/exists and isn't overridden.Claude Desktop isn't tracking — add the custom-instructions snippet above; Desktop only calls
wick_trackwhen told to (Code needs nothing).
Available Tools
5 toolswick_exportA
Export your session data as JSON, CSV, or a Markdown table.
Inputs: format.
wick_resetA
Reset the session counter. Use when starting a new task or after a rate limit reset.
Inputs: no named parameters.
wick_statusA
Get your current Claude session stats — tokens used, cost in USD and INR, burn rate, and messages left estimate.
Inputs: no named parameters.
wick_summaryB
All-time usage summary across every Wick session.
Inputs: no named parameters.
wick_trackA
Track token usage for this response. Call this after EVERY Claude response automatically. Wick uses this to compute your live burn rate, session cost, and estimated messages remaining.
Inputs: model (required), input_tokens (required), output_tokens (required), cache_read_tokens, +1 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 5 tools
Each tool has a clearly distinct purpose: track records usage, status shows current session stats, reset clears the counter, export outputs data, and summary provides all-time totals. There is no ambiguity between them as the descriptions are explicitly differentiated.
All tool names follow a consistent 'wick_' prefix and use lowercase single-word names. However, the second part mixes verbs (track, reset, export) with nouns (status, summary), deviating slightly from a verb_noun pattern, but the overall style is uniform and predictable.
With 5 tools, the server is well-scoped for its purpose of token usage tracking. Each tool serves a necessary function without redundancy, and the count is within the ideal 3-15 range.
The tool set covers the full lifecycle of usage tracking: recording (track), viewing current status (status), resetting sessions (reset), exporting data (export), and summarizing all-time usage (summary). There are no obvious gaps for this domain.
Maintenance
Related MCP Connectors
Hosted MCP server for LLM cost estimation, model comparison, and budget-aware routing.
Shared memory MCP server for Claude Code, Codex, Gemini CLI and Cursor. Cited, editable.
LLM cost optimization gateway, repo token-waste auditor & model arbitrage MCP server.
Hosted Amazon Seller and Vendor MCP server for Claude, ChatGPT, Cursor, Codex, Gemini, Copilot.
Related MCP Servers
- AlicenseAqualityBmaintenanceLocal-first dashboard + MCP server that parses Claude Code and Codex JSONL files into a SQLite cost / token tracker. Per-MCP and per-tool breakdown, session drill-down, dedup by request_id; never talks to vendor APIs5104 npm1MIT
- AlicenseBqualityDmaintenanceTrack LLM token costs across Claude, GPT and Gemini. MCP server + CLI with optimization hints and $ savings estimates.813 npm1MIT
- AlicenseNot gradedqualityDmaintenanceReal-time analytics dashboard for Claude Code, tracking sessions, tool usage, file changes, and token costs via an MCP server.3 npm1MIT
- AlicenseNot gradedqualityDmaintenanceA local MCP server that lets Claude answer plain-language questions about your Anthropic usage and spend across the developer API platform and Claude Enterprise products.MIT