All productsToken Piranha

DEVELOPER TOOLS / AI DIAGNOSTICS ELYRAON STORE

ELYRAON Token Piranha

FREE

Bite the bloat. Keep your tokens. A free local-first AI context overhead scanner that audits Claude Code, Codex, MCP tools, and skills to identify wasted context before you do useful work.

Version 1.0.0. Runs in your browser with no upload, account, payment, or external API.

WHAT IT DOES

Focused help.
Clear boundaries.

Bite the bloat. Keep your tokens. A free local-first AI context overhead scanner that audits Claude Code, Codex, MCP tools, and skills to identify wasted context before you do useful work.

  • Audits Claude Code, Codex, MCP servers, and skills configurations entirely in memory with zero network calls
  • Separates always-loaded per-turn initial overhead from lazily loaded and on-demand assets
  • Pinpoints Token Predators: duplicate instructions, bloated parameter schemas, oversized tool descriptions, and redundant definitions
  • Generates actionable Context Diet optimization plans ranked by token savings and confidence
  • Provides side-by-side Before/After configuration comparison with per-turn impact measurements
  • Exports publication-grade Markdown and machine-readable JSON diagnostic reports
  • Includes a standalone local-first CLI executable for offline terminal workflows

A TYPICAL WORKFLOW

Audit AI coding context bloat

INPUT

Local Claude Code (CLAUDE.md), Codex CLI, MCP tool schemas, or skills directory selected from your filesystem

REVIEWABLE OUTPUT

Instant Context X-Ray report, Feeding Frenzy distribution bars, Token Predators findings, and a prioritized Context Diet action plan.

Workflow summary based on the product documentation. Results depend on the selected input; no customer data is shown here.

token-piranha audit . --local v1.0.0
+-----------------------------------------------------------------------------------+
|  TOKEN PIRANHA v1.0.0  |  "Bite the bloat. Keep your tokens."  |  100% LOCAL SCAN  |
+-----------------------------------------------------------------------------------+
| [CONTEXT X-RAY TELEMETRY]                                                         |
| • Initial Request Turn Overhead:  647 tokens (always-loaded prompt turn tax)      |
| • On-Demand Footprint:             72 tokens (conditionally loaded)               |
| • Total Configured Context:       803 tokens (3,897 bytes across 4 files)         |
| • Potential Turn Savings:        ~274 tokens per turn via Context Diet            |
+-----------------------------------------------------------------------------------+
| [TOKEN PREDATORS IDENTIFIED - 6 Bottlenecks]                                      |
| 1. [HIGH] PREDATOR_DUPLICATE_INSTRUCTION: AGENTS.md & CLAUDE.md (100% match, ~57t)|
| 2. [HIGH] PREDATOR_OVERSIZED_TOOL_DESC: mcp.json [fetch_file_contents] (389 chars)|
| 3. [MEDIUM] PREDATOR_REDUNDANT_TOOL: fetch_file_contents & view_file_stream (69%) |
+-----------------------------------------------------------------------------------+
| [CONTEXT DIET ACTION PLAN]                                                        |
| P1: Deduplicate migration safety instructions between CLAUDE.md & AGENTS.md       |
| P1: Truncate tool descriptions to <140 chars focusing strictly on inputs/outputs    |
| P2: Consolidate overlapping file read tools into single parametrized definition   |
+-----------------------------------------------------------------------------------+