burnlens

    1

    Audit Claude Code token usage, identify high-cost MCP servers, and apply automated fixes to lower API bills.

    $10

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    burnlens

    Example session with this skill installed

    Show me where my tokens went in the last few sessions and help me cut the costs.

    • Read your context and instructions
    • Compiled the burnlens

    I have analyzed your Claude Code transcripts. Total usage reached 1.2M tokens ($18.40 equivalent) with a 42% cache hit rate. The primary cost driver is the `postgres-mcp` server, which accounts for 60% of context overhead due to large schema fetches. ## Token Usage Analysis | Category | Tokens | Cost (Est) | Primary Driver | | :--- | :--- | :--- | :--- | | **MCP Servers** | 720,000 | $10.80 | `postgres-mcp` (schema introspection) | | **Subagents** | 240,000 | $3.60 | `research-agent` (duplicate searches) | | **Tool Calls** | 150,000 | $2.25 | `ls -R` calls in large directories | | **System/Base** | 90,000 | $1.75 | Baseline context | ### Recommended Optimizations 1. **Restrict MCP Scope**: The `postgres-mcp` server is fetching full table schemas on every query. I can update your `.claude/settings.local.json` to include a `table_filter`. 2. **Ignore Directories**: 15% of your tokens are spent indexing `node_modules`. I can add this to your `.claudeignore`. 3. **Subagent Throttling**: The `research-agent` is recursing too deeply. I can modify the `CLAUDE.md` instructions to limit search depth. **Next steps** - Run `python3 "burnlens/scripts/run.py" fixes` to apply the MCP filtering. - Open the full dashboard at `~/.burnlens/dashboard.html` for a per-session breakdown. - Check `savings` in 3 sessions to see the cost reduction impact.

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

    About this skill

    Claude Code token consumption can spiral when MCP servers, subagents, and heavy context accumulation go unchecked. Burnlens parses local session transcripts to pinpoint exactly where your budget is going and provides actionable fixes to reduce overhead.

    What it does

    • Cost breakdown by work type, specific tools, MCP servers, and subagents.
    • Cache analysis to identify hit rates and optimization opportunities.
    • Usage dashboard generation in HTML and plain-text formats for local review.
    • Automated fixes that modify project settings or documentation to trim context.
    • Savings tracking to measure the impact of applied optimizations over time.

    How it works

    1. Analyze usage by running the dashboard command to parse ~/.claude/projects transcripts.
    2. Review insights to identify high-cost tools or subagents via the summary and HTML report.
    3. Execute fixes to generate a list of recommendations for project-level token reduction.
    4. Track savings to compare before and after metrics once new sessions have been processed.

    Frameworks & tools

    This skill requires Python 3.8 or newer. It processes local Claude Code transcripts stored in the standard configuration directories and operates entirely offline.

    Why this beats prompting it yourself

    Raw transcripts are massive and unstructured. Burnlens automates the parsing, categorization, and cost-calculation logic that a manual prompt cannot handle without hitting context limits itself.

    Use cases

    • Identify which MCP server is inflating context windows.
    • Generate a visual breakdown of project spend for cost management.
    • Apply automated rules to .claude/settings.local.json to cap token usage.
    • Validate if recent workflow changes actually lowered your API bill.

    Known limitations

    Requires local access to Claude Code transcript directories. Savings metrics require a few sessions of post-fix usage to calculate accurately.

    Requires Python 3.8+. Only analyzes local Claude Code transcripts. Savings tracking requires multiple sessions to establish a post-fix baseline.

    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

    W3
    Python

    Permissions

    Terminal / Shell
    Browser
    Network Access
    Environment Variables

    Allowed Hosts

    www.w3.org
    Listed1 month ago

    What's inside

    Frequently Asked Questions

    Burnlens is designed specifically for the Claude Code CLI and its associated MCP servers. It parses the local transcripts generated during your coding sessions to provide detailed cost and token analytics.

    The skill identifies inefficient context accumulation and high-cost tool calls, then generates automated fixes for your local project settings. These optimizations help you maintain high performance while significantly reducing your API bill.

    No, Burnlens operates entirely offline by analyzing the session logs stored in your local `~/.claude/projects` directory. Your sensitive code and transcript data never leave your local environment.

    You will receive the Burnlens Python package, documentation for setup, and access to all future updates to ensure compatibility with new versions of the Claude Code CLI.

    Simply ensure you have Python 3.8+ installed, then follow the included instructions to point the skill at your Claude Code configuration directory. The dashboard and analysis commands can be run immediately after installation.

    The skill features a dedicated savings tracking module that compares your token usage metrics from before and after you apply the recommended fixes. You will typically see a clear data trend after 2-3 new coding sessions.

    New in LLM Ops