MCP Server

Active

MIT

Strategic Agent Reasoning MCP MCP Server

Q: What is the Strategic Agent Reasoning MCP?

It's a formal BDI (belief-desire-intention) reasoning engine for autonomous AI agents that provides goal decomposition, multi-attribute utility theory evaluation, belief management with decay, risk assessment, and adaptive replanning. It includes both strategic deliberation tools and a fast System 1 decision layer for sub-2ms tactical decisions.

Q: Is it free?

Yes, it's open-source under the MIT license and available on npm as @putervision/agent-reasoning-mcp.

Q: How do I install it in Cursor or Claude?

Add it to your .cursor/mcp.json or .vscode/mcp.json with the command 'agent-reasoning-mcp' and args ['run']. Set the PENTAD_HMAC_SECRET environment variable for cryptographic token signing.

Q: Does it require authentication or API keys?

It requires a PENTAD_HMAC_SECRET (256-bit shared key) for intention dispatch token signing, but does not depend on external APIs. It can run air-gapped with OFFLINE=1 or SKIP_MODEL_LOAD=1.

Q: What are the performance characteristics?

System 1 fast decision tools target <2ms latency with throughput of 64,000–129,000 ops/s depending on the tool. L1 cache hit latencies are sub-10 microseconds. Full benchmarks are in docs/benchmarks.md.

Q: Can it work offline?

Yes, set OFFLINE=1 or SKIP_MODEL_LOAD=1 to force air-gapped deterministic evaluation using in-memory LRU caches and local heuristics without external model calls.

io.github.putervision/agent-reasoning-mcp

Strategic BDI reasoning engine for autonomous AI agents with goal decomposition, utility theory, and adaptive replanning.

Repository Registry

What is the Strategic Agent Reasoning MCP MCP server?

The Strategic Agent Reasoning MCP is a formal Model Context Protocol server that provides belief-desire-intention (BDI) reasoning, hierarchical goal decomposition, multi-attribute expected utility calculation, and reactive replanning for autonomous AI agents. It combines strategic deliberation tools with a fast System 1 decision layer optimized for sub-2ms latency decisions without external API calls.

This server enables AI agents to reason strategically about goals, evaluate candidate actions using utility theory, manage beliefs with exponential decay, assess risks quantitatively, and adaptively replan when obstacles arise. It's designed for autonomous agents that need formal decision intelligence, explainable reasoning traces, and high-throughput tactical decisions in multi-modal environments.

How to install Strategic Agent Reasoning MCP

Copy-paste configuration for popular MCP clients.

transport: stdio

Config generated by PluginBench — verify against the source before use.

~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "agent-reasoning-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@putervision/agent-reasoning-mcp"
      ]
    }
  }
}

Tools & capabilities

Tools this server exposes to the agent.

set_goal — Manage goal hierarchy, task DAGs, and success criteria with actions: create, update, decompose, get, list, abandon
evaluate_situation — Score and rank candidate actions from environment snapshots with snapshot and quick evaluation modes
replan — Adaptively reconstruct subgoals upon obstacles and abort stale intentions with blocker, event, and full replanning modes
assess_risk — Quantitative threat and risk calculation across candidate actions with action, plan, and compare assessment modes
query_knowledge — Search learned heuristics, tactical knowledge, and past decision patterns via search, patterns, and similar_situations queries
set_utility_weights — Configure utility weights (aggression, caution, greed, efficiency, exploration) with configure, get, list, and activate actions
get_decision_trace — Explainable chain-of-thought rationale and latency telemetry with latest, get, list, and explain modes
manage_beliefs — Structured belief state with exponential confidence decay via update, query, expire, and reconcile operations
manage_intentions — Wire contract directives queue for runtime execution engines with create, dispatch, get, list, cancel, and resolve actions
manage_reasoning_db — Reasoning database statistics, SHA-256 Merkle audit, and snapshot rollback via stats, audit, snapshot, and restore operations
classify — Low-latency categorical labeling over multi-modal state packs with <2ms latency target and ~90,000 ops/s throughput
ask_noul — Typed probabilistic hypothesis and Boolean verification with probability output and <2ms latency
ask_choice — Discrete 1-of-N choice selection (N≤16) with probability simplex and <2ms latency
ask_score — Bounded numeric scalar scoring and calibrated utility rating with <2ms latency
gate_intention — Pre-dispatch blast-radius audit gate issuing signed HMAC dispatch tokens with <1ms latency

Use cases

Autonomous agents decomposing complex multi-step goals into hierarchical subgoals and tracking progress
Evaluating and ranking candidate actions based on multi-attribute utility theory (aggression, caution, efficiency, exploration)
Detecting plan failures and obstacles, then adaptively replanning with new subgoal sequences
Quantitatively assessing risk and threat levels for candidate actions before execution
Making fast tactical decisions (<2ms) using deterministic System 1 heuristics without external API calls

Strategic Agent Reasoning MCP MCP server FAQ

What is the Strategic Agent Reasoning MCP?

It's a formal BDI (belief-desire-intention) reasoning engine for autonomous AI agents that provides goal decomposition, multi-attribute utility theory evaluation, belief management with decay, risk assessment, and adaptive replanning. It includes both strategic deliberation tools and a fast System 1 decision layer for sub-2ms tactical decisions.

Is it free?

Yes, it's open-source under the MIT license and available on npm as @putervision/agent-reasoning-mcp.

How do I install it in Cursor or Claude?

Add it to your .cursor/mcp.json or .vscode/mcp.json with the command 'agent-reasoning-mcp' and args ['run']. Set the PENTAD_HMAC_SECRET environment variable for cryptographic token signing.

Does it require authentication or API keys?

It requires a PENTAD_HMAC_SECRET (256-bit shared key) for intention dispatch token signing, but does not depend on external APIs. It can run air-gapped with OFFLINE=1 or SKIP_MODEL_LOAD=1.

What are the performance characteristics?

System 1 fast decision tools target <2ms latency with throughput of 64,000–129,000 ops/s depending on the tool. L1 cache hit latencies are sub-10 microseconds. Full benchmarks are in docs/benchmarks.md.

Can it work offline?

Yes, set OFFLINE=1 or SKIP_MODEL_LOAD=1 to force air-gapped deterministic evaluation using in-memory LRU caches and local heuristics without external model calls.

README (reference)

Source of truth, from the repository.

@putervision/agent-reasoning-mcp

Strategic BDI Reasoning, Multi-Attribute Expected Utility Theory & Decision Intelligence for Autonomous AI Agents

@putervision/agent-reasoning-mcp is a formal Model Context Protocol (MCP) server that provides strategic belief-desire-intention (BDI) reasoning, hierarchical goal decomposition, multi-attribute expected utility calculation ((E[U] = \sum w_i u_i)), exponential belief decay, quantitative risk evaluation, and reactive replanning across multi-modal memory bridges.

🌐 Official Documentation: putervision.com • Interactive Web Docs

⚡ 15-Second Quick Start

# 1. Initialize reasoning database & seed default utility profiles
npx @putervision/agent-reasoning-mcp init

# 2. Run health diagnostics and Merkle audit checks
npx @putervision/agent-reasoning-mcp doctor

# 3. Inspect active goals, intentions, and belief states
npx @putervision/agent-reasoning-mcp inspect

🛠️ 15 Core MCP Tools

BDI Strategic Deliberation (10 Tools)

Tool	Actions	Purpose
`set_goal`	`create`, `update`, `decompose`, `get`, `list`, `abandon`	Manage goal hierarchy, task DAGs, and success criteria
`evaluate_situation`	`snapshot`, `quick`	Score and rank candidate actions from environment snapshots
`replan`	`blocker`, `event`, `full`	Adaptively reconstruct subgoals upon obstacles and abort stale intentions
`assess_risk`	`action`, `plan`, `compare`	Quantitative threat and risk calculation across candidate actions
`query_knowledge`	`search`, `patterns`, `similar_situations`	Search learned heuristics, tactical knowledge, and past decision patterns
`set_utility_weights`	`configure`, `get`, `list`, `activate`	Configure utility weights (aggression, caution, greed, efficiency, exploration)
`get_decision_trace`	`latest`, `get`, `list`, `explain`	Explainable chain-of-thought rationale and latency telemetry
`manage_beliefs`	`update`, `query`, `expire`, `reconcile`	Structured belief state with exponential confidence decay ($C = C_0 e^{-\lambda t}$)
`manage_intentions`	`create`, `dispatch`, `get`, `list`, `cancel`, `resolve`	Wire contract directives queue for runtime execution engines
`manage_reasoning_db`	`stats`, `audit`, `snapshot`, `restore`	Reasoning database statistics, SHA-256 Merkle audit, and snapshot rollback

System 1 Fast Decision Layer (5 Tools)

Inspired by the typed System 1 pattern pioneered by TypeSafe's Jev (evaluating typed Choice, Score, and Noul primitives over compact state without token generation), implemented locally via in-memory LRU caches and deterministic heuristics (<2ms) without external API calls.

Tool	Purpose	Latency Target	L1 Cache (p50)	Throughput
`classify`	Low-latency categorical labeling over multi-modal StatePacks	`<2ms`	`0.0075 ms`	~90,000 ops/s
`ask_noul`	Typed probabilistic hypothesis and Boolean verification ($p \in [0.0, 1.0]$)	`<2ms`	`0.0049 ms`	~127,000 ops/s
`ask_choice`	Discrete $1$-of-$N$ choice selection ($N \le 16$) with probability simplex	`<2ms`	`0.0138 ms`	~64,000 ops/s
`ask_score`	Bounded numeric scalar scoring and calibrated utility rating	`<2ms`	`0.0057 ms`	~129,000 ops/s
`gate_intention`	Pre-dispatch blast-radius audit gate issuing signed HMAC dispatch tokens	`<1ms`	`0.0709 ms`	~12,000 ops/s

See docs/benchmarks.md for full benchmark reproduction commands, latency percentiles (p50/p95/p99), and multi-tier caching architecture details.

🏛️ PuterVision Pentad Multi-Modal Ecosystem

agent-reasoning-mcp coordinates the closed-loop PuterVision Super-Loop:

🧠 agent-reasoning-mcp: Decides what to do (BDI Strategic Reasoning, Utility Theory, Replanning)
⚡ behavior-mcp: Executes how to act at ~60Hz in browser runtimes
📊 state-memory-mcp: Durable workflow memory, tasks, blockers, decisions
👁️ vision-memory-mcp: Perceptual caching, visual grounding, video timelines
🌐 world-model-mcp: 3D/2D spatial layout, entity permanence, collision simulation

📚 Deep Documentation Guides

📖 Formal API Reference: Full parameter tables, type definitions, and tool schemas for all 15 tools.
🚀 Performance Benchmarks: Empirical throughput and microsecond latency metrics across all 5 System 1 tools.
💡 Core Architecture & Concepts: BDI model, utility formulation, and belief decay dynamics.
🖥️ CLI Usage Guide: Complete CLI command reference (init, doctor, inspect, run).
💾 Database Schema: SQLite table structures, indexes, and Merkle audit ledger.
⚙️ Configuration Reference: .agent-reasoning-mcp.json parameters and environment variables.

🔗 Client Configuration & Environment

Add to .cursor/mcp.json or .vscode/mcp.json:

{
  "mcpServers": {
    "agent-reasoning-mcp": {
      "command": "agent-reasoning-mcp",
      "args": ["run"],
      "env": {
        "PENTAD_HMAC_SECRET": "your-secure-shared-secret-here",
        "DISPATCH_TOKEN_TTL_MS": "30000"
      }
    }
  }
}

Key Environment Variables

PENTAD_HMAC_SECRET: 256-bit shared key for cryptographic intention dispatch token signing.
DISPATCH_TOKEN_TTL_MS: Dispatch token expiration window (default: 30,000ms).
SKIP_MODEL_LOAD: Set to 1 (or OFFLINE=1) to force air-gapped L1/L2 deterministic evaluation.

🧪 Testing & Benchmarks

# Run full unit and integration test suites
npm test

# Run System 1 fast decision layer benchmark suite (throughput & latency percentiles)
npm run benchmark

# Run air-gapped verification
OFFLINE=1 SKIP_MODEL_LOAD=1 npm test

📄 License

MIT © PuterVision

Related MCP servers

Behavior Runtime MCP

Active

High-frequency (~60Hz) in-browser behavior tree execution engine with reactive triggers and safety guardrails for AI agents.

TypeScript

MIT

View repository →

State Memory MCP

Active

Deterministic, persistent SQLite graph for AI agents to track workflow state, decisions, and blockers with zero infrastructure.

TypeScript

MIT

View repository →

Visual Memory MCP

Active

Local visual UI cache for AI agents using perceptual hashing and CLIP to reduce vision token costs.

TypeScript

View repository →

WebCrypt MCP Server

Maintained

Zero-dependency Web Crypto MCP server with AES-256-GCM, RSA-4096, and post-quantum cryptography for AI agents.

JavaScript

MIT

View repository →

World Model MCP

Active

Persistent 3D/2D spatial world model for AI agents with entity tracking, collision simulation, and durable memory.

TypeScript

MIT

View repository →

io.github.putput-io-dev/putput-mcp

Maintained

File uploads for AI agents. Upload, list, and manage files. No signup required.

JavaScript

View repository →