The Local Agent (MCP)
The local agent (MCP)
The SigRank local agent is an MCP that reads your token counts straight from local session logs (22+ platforms supported, including Claude Code, Codex, Amp, Gemini CLI, GitHub Copilot CLI, Goose, Kilo, and more) and keeps your live cascade in sync with the board and your operator profile. You never touch a number, because the agent is the verifier: it counts tokens and never reads the content of your prompts or replies.
What it does
Supported platforms
tokenpull reads local session logs from 22+ AI coding platforms. Each adapter reads that platform's own log format, so you don't reconfigure anything.
“Estimated cache-write” means that platform's log format doesn't expose cache-creation tokens; the other three pillars are exact. Env-var overrides let you point any adapter at a custom log path.
Install
# install globally (recommended)
npm install -g sigrank
# or run without installing
npx sigrank
# wire into Claude Code — .mcp.json
{
"mcpServers": {
"sigrank": { "command": "npx", "args": ["sigrank"] }
}
}In a terminal it opens the TUI, and wired into your AI client it starts the MCP stdio server automatically with no extra config. Verified on Node ≥18, macOS + Linux.
CLI commands
MCP tools — callable by your AI client
When wired into Claude Code or Cursor, your AI agent can call these tools directly, with no paste and no copy-out.
tokenpullOn-device read → 4-window cascade. Zero paste, token-only.tokenpull_submitRead + publish to the board in one call. Server re-scores authoritatively.tokenpull_compareAll four sources side-by-side: tokenpull / ccusage / token-dash / tokscale with delta % per pillar.rank_pasteScore a ccusage / tokscale paste locally. Returns Υ + narration card.rank_windowsScore all four windows from a dashboard paste at once.submit_pasteRank a paste AND publish it to the board in one call.submit_verifiedSign + POST the verified cascade to /api/v1/snapshots (the ranked path).enrollPaste a key from Settings → "New key" → bind this device (signed submit).get_leaderboardLive leaderboard from signalaf.com, any window.get_operatorOne operator's live profile by codename.watch_tokenpullStreaming cascade snapshot — diffs on each poll.diagnose_cascadeDiagnoses where your token cascade is leaking efficiency — ranked findings with severity + estimated Υ impact.simulate_changePrescriptive "what if" — test proposed pillar changes and see the exact Υ delta + class change before committing.suggest_improvementsGenerates ranked, simulated improvement suggestions — tests strategies and returns them sorted by Υ yield impact.self_improveOne-click optimize: diagnoses, suggests, and simulates the best change in a single call.Open by design: the cascade math is public, and proprietary threshold cuts stay server-side. As the canonical anchor, rank_paste reproduces MO§ES Υ 18,436.98 exactly.
How the MCP feeds your operator profile
The agent is the data pipeline between your local session logs and your public operator profile at signalaf.com. Here is the exact path, step by step.
The profile is not separate from the MCP. Every cascade metric your profile displays (Υ Yield, SNR, Leverage, 10xDEV, class tier, per-window history) originates from a tokenpull_submit call (or a manual paste through the calculator), so the profile is the read surface and the MCP is the write path.
The contamination constraint (non-negotiable)
Any live observer that prompts generates the tokens it measures. We learned this directly: a memory observer that auto-prompts (low-input / high-output) inflated a real operator's output by ~25%, visible openly on the live board as the inflated-vs-clean pair (rows 2 and 3). So every SigRank instrument that touches a live session is read-only against telemetry and emits no prompt, with no auto-memory, no keep-alive, and no self-query. It has to be verified-passive, or it re-contaminates every operator running it. This is a hard requirement rather than a caution, and it is the moat: the instrument that doesn't disturb what it measures.
This is the same rule that governs signature drift: the agent is the live reader the drift instrument runs on, and because it never prompts, the drift it reports is the operator's own, not the observer's: a live observer that prompted would inflate the very numbers it reports, and this one cannot.
The agent is how board entries become exact and live, unlike the manual paste calculator, which runs your numbers but does not save to the board or update your profile. Account + review still gate the board so it stays honest.
Build out — the stack
Every piece of technology across the SigRank ecosystem, from the web app to the MCP server to the SIGNOMY marketplace surface.
Token counts only, never prompt content, and verified-passive by design.
FAQ
What is the SigRank local agent?
The SigRank local agent is an MCP server that reads your AI session logs on-device, counts your four token pillars (input, output, cache creation, cache read), and publishes your cascade metrics to the leaderboard. It never reads your prompts or code — only token counts.
Does the SigRank MCP server read my prompts?
No. The SigRank MCP server is read-only and content-free. It counts token quantities (input, output, cache creation, cache read) from your session logs. It never accesses, stores, or transmits your prompts, code, or file contents.
How do I install the SigRank MCP server?
Run npx sigrank to install and start the SigRank MCP server. It works with Claude, Cursor, Cline, Windsurf, and any MCP-compatible client. The local server exposes 30+ tools, including portable TTEOP (Token Telemetry Evaluation Operator Protocol) record export.