Skip to main content
Glama
deenesjakoruzh
mettamatt

Code Reasoning MCP Server

by mettamatt
OverviewSchemaSourceRelated ServersDiscussionsAdmin
TypeScript
Remote

コード推論MCPサーバー

構造化された段階的な思考を通じて複雑なプログラミング タスクを解決する Claude の能力を強化する Model Context Protocol (MCP) サーバー。

npmバージョン ライセンス: MIT CI

クイックインストール

  1. 以下を編集して Claude Desktop を構成します。

    • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

    • Windows: %APPDATA%\Claude\claude_desktop_config.json

    • Linux: ~/.config/Claude/claude_desktop_config.json

    {
  "mcpServers": {
    "code-reasoning": {
      "command": "npx",
      "args": ["-y", "@mettamatt/code-reasoning"]
    }
  }
}

  2. VS Code を構成します。

{
  "mcp": {
    "servers": {
      "code-reasoning": {
        "command": "npx",
        "args": ["-y", "@mettamatt/code-reasoning"]
      }
    }
  }
}

Related MCP server: Sequential Thinking MCP Server

使用法

  1. この MCP をトリガーするには、チャット メッセージに以下を追加します。

    Use sequential thinking to reason about this.

  2. コード推論をトリガーするすぐに使えるプロンプトを使用します。

コード推論プロンプト

  • 特定のコマンドを表示するには、Claude Desktop のチャット ウィンドウの「+」アイコンをクリックするか、Claude Code で/help 。

  • 利用可能なツールから「コード推論から追加」を選択します

  • プロンプトテンプレートを選択し、必要な情報を入力します

  • フォームを送信してチャットメッセージにプロンプトを追加し、リターンキーを押します

プロンプト テンプレートの使用の詳細については、「プロンプト ガイド」を参照してください。

コマンドラインオプション

  • --debug : 詳細なログを有効にする

  • --helpまたは-h : ヘルプ情報を表示する

主な特徴

  • プログラミングの焦点：コーディングタスクと問題解決に最適化

  • 構造化思考：複雑な問題を扱いやすいステップに分解する

  • 思考の分岐：複数の解決策を並行して検討する

  • 思考の修正：理解が深まるにつれて、以前の推論を洗練させる

  • 安全制限: ループを防ぐために 20 回の思考ステップ後に自動的に停止します

  • すぐに使えるプロンプト: 一般的な開発タスク用の事前定義されたテンプレート

ドキュメント

詳細なドキュメントは docs ディレクトリにあります:

プロジェクト構造

├── index.ts                  # Entry point
├── src/                      # Implementation source files
└── test/                     # Testing framework

迅速な評価

コード推論MCPサーバーには、クロードのコード推論プロンプトへの対応能力を評価するプロンプト評価システムが搭載されています。このシステムでは、以下のことが可能です。

  • シナリオ問題に対してさまざまなプロンプトのバリエーションをテストする

  • パラメータ形式の遵守の検証

  • ソリューションの品質を評価する

プロンプト評価システムを使用するには、次のコマンドを実行します。

npm run eval

迅速な比較と開発

Code Reasoningサーバーに最適なプロンプトの開発には多大な労力が費やされました。現在の実装では、評価プロセスを経て最終的に選ばれたHYBRID_DESIGNプロンプトを使用しています。

4 つの異なるプロンプト デザインを比較しました。

プロンプトデザイン

説明

一連

オリジナルのシーケンシャルシンキングプロンプトデザイン

デフォルト

以前サーバーで使用されていたベースラインプロンプト

コード_推論_0_30

コード固有の推論に焦点を当てた実験的な変種

ハイブリッドデザイン

他のアプローチの最良の要素を取り入れた洗練されたデザイン

7 つの異なるプログラミング シナリオにわたる評価では、HYBRID_DESIGN が他のプロンプトよりも優れていることが示されました。

シナリオ

ハイブリッドデザイン

コード_推論_0_30

デフォルト

一連

アルゴリズムの選択

87%

82%

88%

82%

バグの特定

87%

91%

88%

92%

多段階の実装

83%

67%

79%

82%

システム設計分析

82%

87%

78%

82%

コードデバッグタスク

92%

87%

92%

92%

コンパイラの最適化

83%

78%

67%

73%

キャッシュ戦略

86%

88%

82%

87%

平均

86%

83%

82%

84%

HYBRID_DESIGNプロンプトは、平均ソリューション品質（86%）がわずかに最高であり、すべてのシナリオで最も安定したパフォーマンスを示しました（80%を下回るスコアはありませんでした）。また、最も多くの思考を生成しましたsrc/server.tsファイルは、この最適なプロンプトデザインを使用するように更新されました。

個人的には、最大の改善点はプロンプトの最後に次の文を追加したことだと考えています。「✍️ それぞれの考えを「何が欠けているか、または再考する必要があるか」と自問して終わらせます。」

プロンプト評価システムの詳細については、 「テスト フレームワーク」を参照してください。

ライセンス

このプロジェクトはMITライセンスの下で提供されています。詳細はLICENSEファイルをご覧ください。

Available Tools

1 tool
code-reasoningA

🧠 Code Reasoning Tool (using sequential thinking)

Purpose → break complex problems into self-auditing, exploratory thought steps that can branch, revise, or back-track until a single, well-supported answer emerges.

WHEN TO CALL

• Multi-step planning, design, debugging, or open-ended analysis
• Whenever further private reasoning or hypothesis testing is required before replying to the user

ENCOURAGED PRACTICES

🔍 Question aggressively – ask "What am I missing?" after each step
🔄 Revise freely – mark is_revision=true even late in the chain
🌿 Branch often – explore plausible alternatives in parallel; you can merge or discard branches later
↩️ Back-track – if a path looks wrong, start a new branch from an earlier thought
❓ Admit uncertainty – explicitly note unknowns and schedule extra thoughts to resolve them

MUST DO

✅ Put every private reasoning step in thought
✅ Keep thought_number correct; update total_thoughts when scope changes
✅ Use is_revision & branch_from_thought/branch_id precisely
✅ Set next_thought_needed=false only when all open questions are resolved
✅ Abort and summarise if thought_number > 20

DO NOT

⛔️ Reveal the content of thought to the end-user
⛔️ Continue thinking once next_thought_needed=false
⛔️ Assume thoughts must proceed strictly linearly – branching is first-class

PARAMETER CHEAT-SHEET

• thought (string) – current reasoning step
• next_thought_needed (boolean) – request further thinking?
• thought_number (int ≥ 1) – 1-based counter
• total_thoughts (int ≥ 1) – mutable estimate
• is_revision, revises_thought (int) – mark corrections
• branch_from_thought, branch_id – manage alternative paths
• needs_more_thoughts (boolean) – optional hint that more thoughts may follow

All JSON keys must use lower_snake_case.

EXAMPLE ✔️

{
  "thought": "List solution candidates and pick the most promising",
  "thought_number": 1,
  "total_thoughts": 4,
  "next_thought_needed": true
}

EXAMPLE ✔️ (branching late)

{
  "thought": "Alternative approach: treat it as a graph-search problem",
  "thought_number": 6,
  "total_thoughts": 8,
  "branch_from_thought": 3,
  "branch_id": "B1",
  "next_thought_needed": true
}
ParametersJSON Schema
NameRequiredDescriptionDefault
branch_from_thoughtNoBranching point thought number
branch_idNoIdentifier for the current branch
is_revisionNoWhether this is a revision of a previous thought
needs_more_thoughtsNoOptional hint that more thoughts may follow
next_thought_neededYesWhether another thought step is needed
revises_thoughtNoWhich thought is being revised
thoughtYesYour current reasoning step
thought_numberYesCurrent thought number (1-based)
total_thoughtsYesEstimated total thoughts needed (can be adjusted)

TDQS

A4.8/5.0
Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations provided, the description carries the full burden of behavioral disclosure and excels at this. It provides extensive behavioral guidance including 'ENCOURAGED PRACTICES' (questioning, revising, branching, backtracking, admitting uncertainty), 'MUST DO' rules (put every step in thought, keep counters correct, use branching/revision flags precisely, set next_thought_needed=false only when resolved, abort after 20 thoughts), and 'DO NOT' prohibitions (don't reveal thoughts to user, don't continue after next_thought_needed=false, don't assume linear thinking). This comprehensively describes how the tool should be used.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is well-structured with clear sections (Purpose, WHEN TO CALL, ENCOURAGED PRACTICES, MUST DO, DO NOT, PARAMETER CHEAT-SHEET, EXAMPLES) that make it easy to navigate. While comprehensive, it maintains focus with each section serving a clear purpose. Some sections could be slightly more concise, but overall the structure enhances readability and information retrieval.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Given the tool's complexity (9 parameters, no annotations, no output schema), the description provides exceptional contextual completeness. It covers purpose, usage guidelines, behavioral patterns, parameter semantics, and practical examples. The description fully compensates for the lack of annotations and output schema by providing comprehensive guidance on how to use this complex reasoning tool effectively.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema description coverage is 100%, so the baseline is 3. The description adds significant value through the 'PARAMETER CHEAT-SHEET' section that provides practical guidance on parameter usage beyond the schema's basic descriptions. It explains the relationships between parameters (e.g., how is_revision and revises_thought work together, how branching parameters relate) and includes important implementation notes like 'All JSON keys must use lower_snake_case.' The examples further illustrate parameter usage in context.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description clearly states the tool's purpose as 'break complex problems into self-auditing, exploratory thought steps that can branch, revise, or back-track until a single, well-supported answer emerges.' This is specific (verb+resource+methodology) and distinguishes it from any potential alternatives. The 'Purpose →' section provides a concise, accurate summary of what the tool does.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The 'WHEN TO CALL' section explicitly lists scenarios for using this tool: 'Multi-step planning, design, debugging, or open-ended analysis' and 'Whenever further private reasoning or hypothesis testing is required before replying to the user.' It provides clear guidance on when this tool should be invoked versus when to respond directly to the user.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

  1. 1 tool updatev1.0.0
    • First observedcode-reasoning
Deploy Server
One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

TDQS

A4.4/5.0

Scored across 1 tool

Disambiguation5/5

With only one tool, there is no possibility of confusion or overlap between tools. The single tool has a clearly defined purpose for code reasoning and problem-solving, so agents cannot misselect between multiple options.

Naming Consistency5/5

Since there is only one tool named 'code-reasoning', naming consistency is inherently perfect. There are no other tools to compare against, so no inconsistencies can exist in the tool set.

Tool Count2/5

A single tool for a 'Code Reasoning MCP Server' feels too minimal for the apparent scope. While the tool is feature-rich internally, the server's purpose suggests it should offer multiple specialized reasoning tools (e.g., for debugging, design, analysis) rather than one monolithic tool, making the count inappropriate.

Completeness2/5

The server claims to handle 'code reasoning' but provides only one general-purpose tool. This creates significant gaps: there are no specialized tools for different reasoning tasks (e.g., debugging vs. design), no tools for input/output handling, and no way to manage reasoning sessions independently, leading to potential agent failures in complex workflows.

Maintenance

ActivityInactive
ResponsivenessUnresponsive

Resources

Related MCP Connectors

Related MCP Servers

Appeared in Searches