MCP 服务器

Elite Reasoning MCP

A 66-tool reasoning pipeline that intercepts prompts to classify intent, check past mistakes, and generate execution plans, enabling any LLM to think harder and avoid repeating errors.

README

<h3>Make any LLM think harder, reason better, and never repeat mistakes.</h3>

66 tools · Zero API costs · 132KB · Works with any model

<a href="#quick-install">Install</a> · <a href="docs/TOOLS.md">All 66 Tools</a> · <a href="docs/ARCHITECTURE.md">Architecture</a> · <a href="CONTRIBUTING.md">Contributing</a>

<a href="https://cursor.com/install-mcp?name=elite-reasoning&config=eyJjb21tYW5kIjoiYmFzaCIsImFyZ3MiOlsiLWMiLCJjZCB+Ly5lbGl0ZS1yZWFzb25pbmcgMj4vZGV2L251bGwgfHwgZ2l0IGNsb25lIGh0dHBzOi8vZ2l0aHViLmNvbS9TbmVoZ2FiYW5pL2VsaXRlLXJlYXNvbmluZy1tY3AuZ2l0IH4vLmVsaXRlLXJlYXNvbmluZyAmJiBjZCB+Ly5lbGl0ZS1yZWFzb25pbmcgJiYgdXYgcnVuIHB5dGhvbiAtbSBjb3JlLmludGVncmF0aW9uLm1jcF9zZXJ2ZXIiXX0="><img src="https://img.shields.io/badge/Install_in-Cursor-000?style=for-the-badge&logo=data:image/svg+xml;base64,PHN2ZyB3aWR0aD0iMjQiIGhlaWdodD0iMjQiIHZpZXdCb3g9IjAgMCAyNCAyNCIgZmlsbD0ibm9uZSIgeG1sbnM9Imh0dHA6Ly93d3cudzMub3JnLzIwMDAvc3ZnIj48cGF0aCBkPSJNMTIgMkM2LjQ4IDIgMiA2LjQ4IDIgMTJzNC40OCAxMCAxMCAxMCAxMC00LjQ4IDEwLTEwUzE3LjUyIDIgMTIgMnoiIGZpbGw9IiNmZmYiLz48L3N2Zz4=" alt="Install in Cursor" /></a> <a href="https://insiders.vscode.dev/redirect?url=vscode-insiders%3A%2F%2Fanysphere.cursor-mcp%2Finstall%3Fconfig%3DeyJjb21tYW5kIjoiYmFzaCIsImFyZ3MiOlsiLWMiLCJjZCB+Ly5lbGl0ZS1yZWFzb25pbmcgMj4vZGV2L251bGwgfHwgZ2l0IGNsb25lIGh0dHBzOi8vZ2l0aHViLmNvbS9TbmVoZ2FiYW5pL2VsaXRlLXJlYXNvbmluZy1tY3AuZ2l0IH4vLmVsaXRlLXJlYXNvbmluZyAmJiBjZCB+Ly5lbGl0ZS1yZWFzb25pbmcgJiYgdXYgcnVuIHB5dGhvbiAtbSBjb3JlLmludGVncmF0aW9uLm1jcF9zZXJ2ZXIiXX0="><img src="https://img.shields.io/badge/Install_in-VS_Code-007ACC?style=for-the-badge&logo=visual-studio-code&logoColor=white" alt="Install in VS Code" /></a> </div>

The Problem

AI coding assistants make the same mistakes over and over. They forget context between sessions. They don't stress-test their own reasoning. The quality of output varies wildly between prompts.

Elite Reasoning MCP intercepts every prompt and runs it through a reasoning pipeline — classifying intent, checking past mistakes, routing to the right tools, and tracking quality — before the LLM generates a single token.

How It Works

Your Prompt
    │
    ├── Intent Classifier      →  debug / build / design / deploy (13 types)
    ├── Complexity Scorer      →  1–5, adjusts reasoning depth
    ├── Anti-Pattern Checker   →  "You made this mistake before..."
    ├── Prevention Rules       →  Custom auto-triggered safeguards
    ├── MCP & Skill Router     →  Routes to the best available tools
    └── Pre-flight Checklist   →  Per-task reasoning steps
    │
    ▼
Execution Plan → LLM follows it → Better output

Everything runs locally. No API calls. No cloud. A single SQLite file stores anti-patterns, decisions, quality scores, and calibration data across sessions.

Quick Install

pip (Recommended)

pip install elite-reasoning-mcp

From Source (macOS / Linux)

git clone https://github.com/Snehgabani/elite-reasoning-mcp.git ~/.elite-reasoning
cd ~/.elite-reasoning && bash scripts/install.sh

Windows (PowerShell)

git clone https://github.com/Snehgabani/elite-reasoning-mcp.git $env:USERPROFILE\.elite-reasoning
cd $env:USERPROFILE\.elite-reasoning; .\scripts\install.ps1

Docker

docker run -v elite-brain:/data/brain ghcr.io/snehgabani/elite-reasoning-mcp

Requirements: Python 3.11+ and uv (auto-installed by the installer).

What's Inside

66 Tools, 7 Categories

Category	Tools	What They Do
Core Pipeline	`orchestrate_request_tool` `assess_confidence` `reasoning_preflight`	Intent classification, complexity scoring, pre-flight checklists
Quality & Memory	`record_mistake` `check_anti_patterns` `record_quality_score` `get_quality_trend`	Anti-pattern database, quality tracking, trend analysis
Decision Making	`decision_council_review` `record_decision` `search_decisions`	5-perspective adversarial review, decision logging
Risk Analysis	`fmea_analysis` `swiss_cheese_audit` `smoke_test_gate` `bias_scan`	Failure mode analysis, defense layer audit, bias detection
Calibration	`calibration_predict` `calibration_resolve` `calibration_score`	Prediction tracking with Brier scores
Learning	`record_hypothesis` `five_whys` `after_action_review` `socratic_challenge`	Root cause analysis, hypothesis testing, self-challenge
Autonomous	`autonomous_scan` `self_diagnose` `predictive_prevention`	Self-monitoring, proactive issue detection

Full reference: docs/TOOLS.md

Key Capabilities

Anti-Pattern Memory

Stores every mistake with root cause and fix. The LLM checks this database before acting — so it never makes the same mistake twice.

</td> <td width="50%">

Decision Council

Five adversarial perspectives review every major decision: Security, Scalability, Simplicity, User Impact, and Future Self.

</td> </tr> <tr> <td width="50%">

Confidence Calibration

Tracks prediction accuracy with Brier scores. You learn when to trust the LLM's confidence — and when to doubt it.

</td> <td width="50%">

Cross-Session Memory

Knowledge persists in a local SQLite database. Context compounds across conversations instead of resetting.

</td> </tr> <tr> <td width="50%">

FMEA Risk Analysis

Failure Mode and Effects Analysis before you build. Catch what can go wrong before it does.

</td> <td width="50%">

Custom Prevention Rules

Auto-triggered safeguards for your workflow. "Always check X before doing Y" — enforced automatically.

</td> </tr> </table>

IDE Support

Works with any MCP-compatible client via stdio transport:

IDE	Config File
Cursor	`~/.cursor/mcp.json`
Claude Desktop	App Settings → MCP
VS Code + Continue	`.continue/config.json`
Windsurf	`~/.codeium/windsurf/mcp_config.json`
Antigravity	`~/.gemini/config/mcp_config.json`

<details> <summary>Manual Configuration</summary>

Add to your MCP config:

{
  "mcpServers": {
    "elite-reasoning": {
      "command": "bash",
      "args": ["-c", "cd ~/.elite-reasoning && uv run python -m core.integration.mcp_server"]
    }
  }
}

On Windows, use scripts/run_elite_mcp.bat instead:

{
  "mcpServers": {
    "elite-reasoning": {
      "command": "cmd",
      "args": ["/c", "%USERPROFILE%\\.elite-reasoning\\scripts\\run_elite_mcp.bat"]
    }
  }
}

</details>

Architecture

elite-reasoning-mcp/
├── core/
│   ├── integration/
│   │   └── mcp_server.py          # FastMCP server, tool registration
│   ├── memory/
│   │   ├── persistent_store.py    # SQLite — 15 tables, 32 indexes
│   │   ├── graph_store.py         # Knowledge graph
│   │   └── embedding.py           # Semantic search (optional)
│   ├── tools/
│   │   ├── orchestration.py       # Intent classification & routing
│   │   ├── reasoning_amplifier.py # Calibration, council, preflight
│   │   ├── adaptive.py            # Learning & user modeling
│   │   ├── analysis.py            # Risk, FMEA, confidence
│   │   ├── auditing.py            # Quality & anti-patterns
│   │   └── planning.py            # Goals & benchmarks
│   └── identity/
│       └── user_profile.py        # Per-user configuration
├── schemas/                       # 66 JSON tool schemas
├── scripts/                       # Installers & launchers
└── Dockerfile                     # Container support

Full architecture docs: docs/ARCHITECTURE.md

Performance

Metric	Value
Package size	132KB
Startup time	< 2s
Per-tool latency	< 50ms
Storage	Local SQLite
API costs	$0
Network calls	Zero

FAQ

<details> <summary>Does this work with weak / open-source models?</summary> Yes. The pipeline runs before the model generates output, so even weaker models get structured reasoning plans, anti-pattern checks, and quality tracking. The tools amplify whatever model you're using. </details>

<details> <summary>Will it slow down my responses?</summary> No. Each tool call takes < 50ms. The orchestrator adds one tool call per prompt. Total overhead is negligible. </details>

<details> <summary>How is this different from sequential-thinking?</summary> Sequential-thinking gives the LLM a scratchpad for multi-step reasoning. Elite Reasoning MCP goes further: it classifies intent, checks past mistakes, routes to specialized tools, tracks quality over time, and builds persistent memory across sessions. They're complementary — use both. </details>

<details> <summary>Is my data private?</summary> 100%. Everything runs locally. The SQLite database is on your machine. No telemetry, no API calls, no cloud. Your code and prompts never leave your computer. </details>

Contributing

We welcome contributions. See CONTRIBUTING.md for guidelines.

Good first issues:

Add new reasoning tools
Improve the intent classifier
Write tests
Add benchmarks for output quality with/without the pipeline

Security

All data stays local. Zero network calls. No telemetry. See SECURITY.md for the full policy.

Found a vulnerability? Email snehgabani@users.noreply.github.com — don't open a public issue.

License

MIT — use it however you want.

<div align="center"> Built by <a href="https://github.com/Snehgabani">@Snehgabani</a> <a href="https://github.com/Snehgabani/elite-reasoning-mcp"> <img src="https://img.shields.io/badge/Star_on_GitHub-★-yellow?style=flat-square" alt="Star" /> </a> </div>

Elite Reasoning MCP

README

The Problem

How It Works

Quick Install

pip (Recommended)

From Source (macOS / Linux)

Windows (PowerShell)

Docker

What's Inside

66 Tools, 7 Categories

Key Capabilities

IDE Support

Architecture

Performance

FAQ

Contributing

Security

License

推荐服务器