More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Mcp Tax

    1

    Measure how many tokens your MCP servers burn before you type, then get a concrete plan to reclaim the space.

    $19

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    mcp-tax

    mcp-tax

    Example session with this skill installed

    my claude code sessions start degrading way too fast and i have a bunch of mcp servers installed. audit what they are costing me in context and tell me what to actually do about it.

    • Read your context and instructions
    • Compiled the mcp-tax

    mcp-tax digest

    Total: ~29,596 tokens across 8 live servers
    = 14.8% of a 200,000 window
    = 23.1% of a 128,000 window
    = 92.5% of a 32,000 window

    servertools~tokenssharebiggest tool
    claude-code2014,32448.4%Bash (2543)
    chrome-devtools325,88419.9%emulate (569)
    blender93,52611.9%look (718)
    hostinger-vps141,5805.3%read_text_file (180)
    server-31-97-202141,5805.3%read_text_file (180)
    local-filesystem141,5805.3%read_text_file (180)
    browsermcp128212.8%browser_type (124)
    brave-search23011.0%brave_web_search (154)

    Biggest single tools

    • claude-code/Bash: 2,543 tokens (86% is the description text)
    • claude-code/SendMessage: 1,673 tokens (68% is the description text)
    • claude-code/Agent: 1,597 tokens (72% is the description text)

    Reclaim plan

    1. claude-code alone is 48.4% of the total. Gate it to the projects that need it via .mcp.json and reclaim ~14.3k tokens everywhere else.
    2. Eleven claude-code tools carry bloated descriptions (Bash's description is 2,179 of its 2,543 tokens). Trimming upstream frees ~5.4k more.
    3. Three servers are near-duplicates of the same filesystem spec at 1,580 each. Keep one, reclaim ~3.2k.

    Net effect of acting on all three: roughly 19k of 29.6k tokens back, about 9% more usable window on a 200k context.

    Counts approximate the serialized payload; real cost varies by client and tokenizer.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Every tool, prompt and resource an MCP server exposes is serialized into your model's context window on every turn. Add a few servers and the meter is already running: a 30k token MCP setup on a 128k window leaves under 100k for the actual conversation.

    mcp-tax measures that cost against your real configs (Claude Desktop, Claude Code, Cursor, project .mcp.json, .vscode/mcp.json) by spawning each server over stdio exactly like your client does, then produces a ranked digest: tokens per server, the biggest single tools, how much of each tool's cost is description text vs schema, and your total as a share of 200k / 128k / 32k windows.

    The skill then turns the digest into a reclaim plan with estimates per action: which servers to gate per-project, which tool descriptions are bloated, and where the savings actually are.

    What's inside

    • scripts/audit.mjs runs mcp-tax-cli over your live configs and prints a ranked digest with reclaim estimates
    • reference/context-budget.md is the playbook: per-project gating, description trimming, near-duplicate servers, on-demand heavy servers
    • Works with Claude Code, Cursor, Codex CLI, Gemini CLI and other SKILL.md-compatible agents

    Honest limits

    • Token counts approximate the serialized payload; real cost varies by client and tokenizer
    • It measures the static manifest. Tools a server reveals only at runtime are invisible to it
    • Requires Node.js 18+ and permission to spawn your configured servers locally

    Requires Node 18+. Reports are approximations of serialized payloads. Cannot measure servers that fail to start over stdio.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    Requires

    Tax Cli MCP
    Node.js

    Permissions

    Write Files
    Network Access
    Read Files
    Terminal / Shell
    Listed1 month ago

    What's inside

    Frequently Asked Questions

    This skill identifies 'context rot' caused by MCP servers that silently consume your token limit with bloated tool descriptions, helping you reclaim space for actual conversation and reducing agent sluggishness.

    The skill is compatible with any environment using the Model Context Protocol, including Claude Desktop, Claude Code, Cursor, and custom projects utilizing local .mcp.json configurations.

    You will receive the full skill package, including the audit agent instructions, access to the mcp-tax-cli tool for local execution, and a specialized context budget playbook for optimization strategies.

    Setup is minimal; as long as you have Node 18+ installed, the skill uses npx to interface with your existing MCP configurations without requiring manual server re-installation.

    Yes, the audit script is designed to scan both global configurations (like Claude Desktop) and project-specific .mcp.json files to provide a comprehensive view of your total token overhead.

    New in LLM Ops