MisakaNet is an agent failure memory network with 235+ verified debugging lessons. It provides 3 MCP tools (search, get_lesson, submit_usage), 5 resources, and 3 guided prompts for failure triage. Works with Claude Code, Cursor, and any MCP client.
Use ontorata/ratary Glama badge instead of mindmap badge. Restore full mindmap-mcp-server listing to resolve duplicate-check bot feedback on PR #9454.
Co-authored-by: Cursor <cursoragent@cursor.com>
Adds asset-management to Finance & Fintech:
A local, read-only MCP server over your own stock/ETF transaction log. A deterministic core computes drawdown-first risk (max drawdown, Ulcer, CDaR, Sharpe/Sortino/Calmar) with bootstrap CIs and walk-forward verdicts that say "inconclusive" when evidence is thin — reconciled to the cent against a real broker export.
The AI is kept off the numbers in two layers, with different (and honestly-scoped) guarantees:
- MCP server (this listing): the model calls read-only tools that return core-computed figures; it's instructed to report those and never compute its own, and a portfolio://guarantees resource states the contract. Provision + instruction — not a mechanical block, since the chat client isn't ours.
- CLI narration (optional): a by-construction fence — the model writes only {{token}} placeholders, a renderer substitutes validated figures, and one model-typed digit voids the whole note. Here the model literally cannot emit a number.
Python 3.12 / uv, stdio, AGPL-3.0.
Base token-safety MCP toolkit: the only B20 (Base-native token)
freeze/seize/rug suite, plus pre-trade gates, honeypot/sellability,
wallet + approval audits, sign-guard, prices and AI reports. 80+
pay-per-call tools over x402, free tier, no API keys.