Early access — cascade metrics are real (derived from canonical token telemetry); the operator field is a curated seed. Learn more about the data

SigRank MCP Server

16 remote tools (HTTP, no install) + 25 local tools (npx sigrank) — measure, rank, benchmark, compare, and improve token efficiency. The yield cascade and live leaderboard as MCP tools.

Remote endpoint (HTTP, no install)

https://signalaf.com/api/mcp

Streamable HTTP transport. Protocol 2025-06-18. No auth required. Discovery card at /.well-known/mcp.json. Works with Cloudflare AI Playground, web agents, and any MCP client that supports Streamable HTTP.

Remote tools (16)

get_sigrank_standard_recordReturn a portable sigrank/0.1-draft record from I/O/W/R telemetry, with only the five base metrics
rank_pasteCompute cascade metrics (Υ, SNR, Leverage, Velocity, 10xDEV) from 4 token counts — no install needed
get_leaderboardPull the live global leaderboard from signalaf.com
get_operatorGet any operator's profile, class tier, and cascade breakdown
simulate_changeWhat-if predictor — test pillar changes and see the exact Υ delta before committing
diagnose_cascadeIdentify efficiency leaks with severity, findings, and estimated Υ impact per fix
suggest_improvementsRanked, simulated improvement strategies sorted by Υ yield impact
self_improveOne-click optimize — diagnose → suggest → simulate the best change in one call
rank_windowsScore 7d / 30d / 90d / all-time windows in one call
benchmark_meAnswers 'How good am I?' — benchmarks your cascade against the live field: percentile, rank, strongest/weakest metric, one-line interpretation
rank_ifCounterfactual rank simulator — 'What would it take to reach top 10%?' — finds the smallest pillar change to hit a target percentile
operator_gapWhat specifically separates two operators — primary cause, secondary cause, offsetting weakness, most explanatory factor
field_anomalyUnusual patterns in the leaderboard — high velocity with low leverage, near-zero cache write in top 50, largest 7d improvements, extreme divergence
who_operates_like_meNearest-neighbor finder — operators whose signature most resembles yours, with similarity % and where they outperform you
compare_to_fieldYOU vs FIELD vs TOP 10% vs TOP 1% comparison table for yield, leverage, velocity, and SNR
operator_signaturePortable identity object — signature code, archetype, dominant trait, and closest comparable operators

Resources (6)

sigrank://methodologyHow SigRank measures AI operators — the cascade metric system, formulas, and class taxonomy
sigrank://metricsDefinitions of Yield (Υ), Leverage, Velocity, SNR, 10xDEV, Scale V, and class tiers
sigrank://platformsAI platforms tracked by SigRank
sigrank://formulasThe frozen canonical formulas — Υ, Leverage, Velocity, SNR, 10xDEV, Scale V
sigrank://classesThe 24-stage class taxonomy from IGNITER III to ARCH+ I with token thresholds
sigrank://benchmarksCurrent field-wide benchmark statistics (median, top 10%, top 1%) from the live leaderboard

Prompts (5)

benchmark-my-operatorCompute cascade metrics, compare against the live field, and get a one-line interpretation
how-do-i-reach-top-10Counterfactual analysis — finds the smallest pillar change needed to reach a target percentile
explain-my-signatureComputes operating archetype, dominant trait, and finds comparable operators
diagnose-inefficiencyIdentifies efficiency leaks with severity, findings, and estimated yield impact per fix
field-anomaly-reportScans the live leaderboard for unusual patterns — no input required

Local install (stdio, 24 tools)

npx sigrank

Or npx sigrank-mcp — same thing. Bun users: bunx sigrank. No API key required for read tools. Submit tools need a Supabase anon key (free, set during sigrank enroll). The local server reads session logs from 17 AI coding tools (Claude Code, Codex, Devin, Amp, Kimi, pi-agent, oh-my-pi, OpenClaw, Droid, Codebuff, Hermes, Kilo, Qwen, Goose, Gemini, Copilot, OpenCode).

Client setup (local stdio)

ClientWhere
Claude Desktop~/Library/Application Support/Claude/claude_desktop_config.json
CursorSettings → MCP → Add Server
ClineMarketplace → search "SigRank"
WindsurfSettings → MCP Servers → Add
Any MCP clientnpx sigrank or bunx sigrank (stdio transport)

Local tools (25)

get_sigrank_standard_recordReturn the portable SigRank Standard record for four token counts
rank_pastePaste token counts, get projected SigRank rank in seconds
get_leaderboardPull the live global leaderboard from signalaf.com
get_operatorGet any operator's profile, class tier, and cascade breakdown
submit_pasteSubmit token telemetry to the live board (paste mode)
submit_verifiedSubmit ed25519-signed telemetry (verified mode)
tokenpullPull telemetry from local session logs (17 platform adapters)
tokenpull_submitPull telemetry and submit in one call
watch_tokenpullContinuous monitoring with periodic submissions
rank_windowsCompare rank across 7d / 30d / 90d / all-time windows
tokenpull_compare4-source pillar audit (tokenpull vs ccusage vs token-dashboard vs tokscale)
enrollRegister as an operator on the leaderboard
simulate_changeModel what-if scenarios (what if cache hit rate +10%?)
diagnose_cascadeIdentify inefficiencies in token usage patterns
suggest_improvementsGet actionable recommendations to improve yield
self_improveFull cycle: diagnose → suggest → simulate, with mode detection and scoped analysis
discover_peersFind mentors, peers, and complementary operators for your operator
tokscale_breakdownPer-model token breakdown across platforms
tokscale_market_shareAI tool market share from local tokscale data
tokscale_developer_profilePer-developer usage profile (model mix, sessions, workspaces)
tokscale_model_trendsModel adoption over time (first/last seen, monthly curve)
tokscale_cost_analysisCost per developer per model
tokscale_device_profileDevice fingerprinting (installed tools, session counts, day-of-week)
tokscale_mcp_usageMCP server usage (detected servers, active days)
tokscale_competitive_intelCompetitive intelligence for any AI tool

Find us on

Try the score calculator →View the leaderboard →GitHub repo →npm →