Hokmah

io.github.davidangularme由 GitHub 帳號 davidangularme 發布v1.0.0更新於 Oct 24, 2026

AI Agent with Architectural Memory. Impact analysis (free), tests and code from the graph (pro).

已驗證Streamable HTTP可網頁執行知識與記憶開發者工具

概覽

AI 產生的概覽

Hokmah 是一個遠端 MCP 伺服器，可將公開 GitHub 儲存庫匯入架構圖，讓助理評估所描述程式碼變更的影響。

可以做什麼
匯入公開 GitHub 儲存庫，將其原始檔與近期提交索引為相依性與變更歷史圖；branch 參數僅為相容而接受，實際使用預設分支。評估所描述程式碼變更對已匯入儲存庫的影響，回傳風險分數、受影響檔案與架構不變量。連接外部 MCP 伺服器以進行編排。README 另列出專業版從圖中產生測試與程式碼，但這些不在本清單宣告的工具中。
適用情境
當你希望助理在不把整個程式碼庫送給模型的情況下，評估公開 GitHub 儲存庫中某次重構或變更的影響範圍，並希望跨工作階段保留該儲存庫的架構檢視時適用。
執行需求
需要遠端 streamable HTTP 端點，無需本機執行環境或安裝套件。儲存庫必須在 GitHub 上公開。託管端點未宣告身分驗證。自架需要 Python、上游 Hokmah 後端，以及 HOKMAH_API_BASE、HOKMAH_MCP_PORT 和 HOKMAH_PRO_KEYS 設定。
安裝前請注意
匯入儲存庫會將其原始檔與提交歷史傳送到第三方託管服務。README 說明了付費專業版，使用你自己的 LLM 金鑰產生測試和程式碼，並需向廠商取得 Pro 金鑰；宣告的工具不包含這些。分析結果來自圖模型的估算，並非經過驗證的測試結果。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 Hokmah，將其新增到工作區。
  2. 為需要使用其工具的對話啟用該服務。

Web executable，透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。

交給代理程式安裝

把這段提示詞貼上到 Claude Code、Codex、Cursor 等代理程式，或 SourceWeft 對話中。代理程式會讀取該伺服器的安裝指南，先向你展示發布者、端點和登入方式，經你同意後再把它新增到你的客戶端。金鑰不會寫進任何命令。

閱讀 https://sourceweft.com/mcp/io.github.davidangularme%2Fhokmah/install.md，並按其中說明把這個 MCP 伺服器接入我的客戶端。修改任何設定之前，先告訴我它的發布者、端點和登入方式，等我確認。不要把 API 金鑰寫進命令或我會提交的檔案；問我金鑰應從哪裡取得。

檢視代理程式所依據的安裝指南

其他 MCP 客戶端

把它新增到你客戶端的 mcpServers 設定中。

{
  "mcpServers": {
    "hokmah": {
      "type": "http",
      "url": "https://hokmah.dev/mcp"
    }
  }
}

README

Hokmah MCP Server

AI Agent with Architectural Memory — MCP Server

Gives any AI coding agent persistent understanding of codebases via TransitionGraph, IdeaGraph, and WorldModel. Analyze impact, generate tests, write code — all from the graph.

Quick Start (30 seconds)

Add to your editor's MCP config (Cursor, Claude Code, VS Code, Windsurf, Cline, JetBrains):

json
{  "mcpServers": {    "hokmah": {      "type": "streamable-http",      "url": "https://hokmah.dev/mcp"    }  }}

Then ask your agent: "analyze the impact of refactoring the auth module in github.com/owner/repo"

Available Tools

ToolTierDescription
hokmah_analyzeFREEImpact analysis, risk score, affected files, architectural invariants
hokmah_connect_projectFREEConnect a GitHub repo, build the architectural graph
hokmah_connect_mcpFREEConnect an external MCP server for orchestration
hokmah_generate_testsPROTest generation from the graph (40x fewer tokens)
hokmah_generate_codePROCode generation with architectural memory

How It Works

Hokmah builds a persistent architectural graph from your codebase:

  • TransitionGraph — Markov model of code changes (which files change together)
  • IdeaGraph — 16 relation types between concepts
  • WorldModel — File tree, dependencies, symbols

When you ask "what's the impact of changing X?", Hokmah traverses the graph instead of sending your entire codebase to an LLM. That's why analyze is free (zero LLM tokens) and generate uses 40x fewer tokens.

Pricing

  • Free — hokmah_analyze + hokmah_connect_project + hokmah_connect_mcp (unlimited)
  • Pro — hokmah_generate_tests + hokmah_generate_code (BYOK — bring your own LLM key)

Get a Pro key at hokmah.dev.

Editor Setup

  • Cursor — Settings → MCP → Add server → paste config
  • Claude Desktop — ~/Library/Application Support/Claude/claude_desktop_config.json
  • Claude Code — claude mcp add hokmah --transport streamable-http --url https://hokmah.dev/mcp
  • VS Code (Copilot) — .vscode/mcp.json in project root
  • Windsurf — ~/.windsurf/mcp.json
  • Cline — Settings → MCP Servers → Add
  • JetBrains — Settings → Tools → AI Assistant → MCP Servers

Self-Hosting

The hosted server at https://hokmah.dev/mcp is the recommended way to use Hokmah. To run the server yourself against your own Hokmah backend:

bash
pip install -r requirements.txtcp pro_keys.example.json pro_keys.json  # edit with your real PRO keysHOKMAH_API_BASE=http://localhost:8000 python mcp_server.py

Environment variables:

  • HOKMAH_API_BASE — upstream Hokmah API (default http://localhost:8000)
  • HOKMAH_MCP_PORT — port to listen on (default 8001)
  • HOKMAH_PRO_KEYS — path to the PRO keys JSON file (default /home/vpm/mcp-server/pro_keys.json)

A reference systemd unit is provided in hokmah-mcp.service.

Built by

Catalyst AI Research · Haifa, Israel

License

MIT

來源：README.md，提交 78d5b00

工具

2

hokmah_analyze

Analyze Change Impact

Estimates the impact of a described code change on an imported public GitHub repository. `repo_url` is https://github.com/<owner>/<repository> and `query` describes the change in plain language; `branch` is accepted for compatibility, the analysis uses the history imported from the default branch. Returns a risk assessment with the files, symbols and modules likely to be affected and related commits, computed from the repository graph without a language model.

輸入參數query必填branchrepo_url必填

hokmah_connect_project

Import Public GitHub Repository

Imports a public GitHub repository so it can be analyzed: indexes its source files (.py .js .ts .jsx .tsx .rs .go .java) and its latest commits (up to 500) from the default branch into a dependency and change-history graph. `repo_url` is https://github.com/<owner>/<repository>; `branch` is accepted for compatibility. A repository already imported is not imported again. Returns the project id and the number of commits and files ingested.

輸入參數branchrepo_url必填

版本歷史

1
  1. v1.0.0最新Sep 16, 2026

評分

已滿足適用檢查項的 100%

該評分衡量此收錄條目的完整程度，依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價，也不表示伺服器安全或目前可用：可連通性是帶日期的檢查結果，評分也不包含安裝量或使用者評價。

  • 近 30 天內可連通必需25 分
  • 已宣告認證方式必需10 分
  • 已列出工具必需15 分
  • 描述必需15 分
  • 圖示10 分
  • 文件15 分
  • 授權條款10 分