OraClaw -- 12 MCP tools for local AI decision-making (optimization, forecasting, risk)

[Whatsonyourmind]

Hey LocalAI community! Built an MCP server that adds deterministic decision-making tools to local AI setups.

Why this matters for local AI

When running models locally, every token counts. OraClaw offloads mathematical decisions (A/B testing, scheduling, risk calculation, forecasting) to lightweight, deterministic algorithms that run in sub-25ms -- letting your local model focus on reasoning rather than computation.

Setup

npx @oraclaw/mcp-server

Or call the REST API directly:

curl -X POST https://oraclaw-api.onrender.com/api/v1/optimize/bandit \
  -H "Content-Type: application/json" \
  -d '{"arms": [{"id":"A","pulls":100,"totalReward":72},{"id":"B","pulls":50,"totalReward":40}], "algorithm":"ucb1"}'

The 12 MCP tools

Tool	Algorithm	Latency
optimize_bandit	UCB1, Thompson, Epsilon-Greedy	0.01ms
optimize_contextual	LinUCB	0.02ms
optimize_cmaes	CMA-ES	12ms
solve_constraints	LP/MIP via HiGHS	1ms
solve_schedule	Energy-matched scheduling	0.5ms
analyze_decision_graph	PageRank, Louvain	0.05ms
analyze_portfolio_risk	VaR, CVaR	0.03ms
score_convergence	Multi-source agreement	0.01ms
score_calibration	Brier, log score	0.01ms
predict_forecast	ARIMA, Holt-Winters	0.5ms
detect_anomaly	Z-Score, IQR	0.01ms
plan_pathfind	A*	0.1ms

All pure TypeScript, no GPU needed. MIT licensed, 1,072 tests.

Links: GitHub | npm | Demo