Melchizedek

io.github.louis49由 GitHub 账号 louis49 发布v1.0.2更新于 Oct 25, 2026

Persistent memory for Claude Code — offline, single-file, hybrid search.

已验证STDIO仅桌面知识与记忆开发者工具数据与分析

概览

AI 生成的概览

本地 MCP 服务器，把 Claude Code 对话记录索引到单个 SQLite 文件，让助手跨项目检索历史会话。

可以做什么
检索已索引的对话，返回精简片段，支持日期过滤和排序，并可继续获取上下文或完整内容逐层深入。查找涉及某个文件的过往会话、某条报错信息的历史解决方案，或类似任务的以往做法。手动保存记忆笔记，列出已索引会话，可按项目筛选。查看索引统计与配置信息，并修改设置。从索引中删除单个片段或整个会话，排除或重新纳入项目，以及重启服务器。
适用场景
适合在 Claude Code 中跨多个会话和项目工作、希望自动回忆以往调试过程、架构决策和代码模式的用户，无需反复说明背景。也适合要求这些历史完全留在本机的用户。
运行要求
以 npm 包形式通过 stdio 在本机运行，需要 Node.js 20 或更高版本、Claude Code 2.0 或更高版本，支持 macOS、Linux 或 Windows。自动索引需要在 Claude Code 设置文件中添加 SessionEnd 钩子。可选的向量与重排模型会在首次使用时下载。
安装前请注意
它会读取 Claude Code 对话记录并建立可检索的本地索引，因此对话中的敏感内容除非放入 private 标签，否则都可被检索到。删除与遗忘类工具会永久移除索引数据，清除被忽略项目的数据后无法恢复。可选模型下载会产生网络请求，其余均在本机完成。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Melchizedek，将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Desktop only，通过 STDIO。 STDIO 服务会启动本地进程，因此需要 SourceWeft 桌面宿主。

其他 MCP 客户端

参照 仓库 中的启动说明。

README

Melchizedek

[npm version] [npm downloads] [CI] [License: MIT] [Donate] [Donate]

Persistent memory for Claude Code. Automatically indexes every conversation and provides production-grade hybrid search (BM25 + vectors + reranker) via MCP tools. 100% local, zero config, zero API keys, zero invoice.

Why Melchizedek?

Claude Code forgets everything between sessions - and knows nothing about your other projects. Melchizedek fixes both.

It runs silently in the background - indexing your conversations as you work - then gives Claude the ability to search across your entire history, across all projects: past debugging sessions, architectural decisions, error solutions, code patterns.

No cloud. No API keys. No config. Plug and ask.

How it works

~/.claude/projects/**/*.jsonl       (your conversation transcripts - read-only)        |        v  SessionEnd hook                   (auto-triggers after each session)        |        v  +-----------------+  |  Indexer         |    Parse JSONL -> chunk pairs -> SHA-256 dedup  |  (better-sqlite3)|    FTS5 tokenize -> vector embed (optional)  +-----------------+        |        v  ~/.melchizedek/memory.db           (single SQLite file, WAL mode)        |        v  +-----------------+  |  MCP Server      |    16 search & management tools  |  (stdio)         |    Hybrid: BM25 + vectors + RRF + reranker  +-----------------+        |        v  Claude Code                       (searches your history via MCP)

Search pipeline - 4 levels of graceful degradation

Every layer is optional. The plugin works with BM25 alone and gets better as more components are available.

LevelComponentWhat it addsDependency
1BM25 (FTS5)Keyword search with stemmingNone (always active)
2Dual vectors (sqlite-vec)Semantic search - text (MiniLM 384d) + code (Jina 768d)@huggingface/transformers (optional)
3RRF fusionMerges BM25 + text vectors + code vectors via Reciprocal Rank FusionVectors enabled
4RerankerCross-encoder re-scoring of top resultsTransformers.js or node-llama-cpp (optional)

Performance

Measured with npm run bench - 100 sessions, 1 000 chunks, on a single SQLite file.

MetricResultTarget
Indexation (100 sessions)~80 ms< 10 s
BM25 search (mean)~0.2 ms< 50 ms
DB size (100 sessions)~1.4 MB< 30 MB
Tokens per search~125< 2 000

Quick Start

npm (recommended)

bash
npm install -g melchizedek

Add the MCP server to Claude Code:

bash
claude mcp add --scope user melchizedek -- melchizedek-server

npx (no install)

bash
claude mcp add --scope user melchizedek -- npx melchizedek-server

From source

bash
git clone https://github.com/louis49/melchizedek.gitcd melchizedek && npm install && npm run buildclaude --mcp-config .mcp.json

Claude Code plugin marketplace (coming soon)

Plugin review pending. In the meantime, use npm or npx install above.

bash
claude plugin install melchizedek   # not yet available

Setting up hooks (automatic indexing)

The MCP server provides search tools, but hooks trigger automatic indexing. Without hooks, you'd need to manually index sessions.

For marketplace installs, hooks are configured automatically. For npm/npx/source installs, add hooks to ~/.claude/settings.json.

See docs/installation.md for the full JSON configuration, hook reference, and troubleshooting.

After setup, restart Claude Code. Indexing starts automatically.

MCP Tools

Search (start here)

ToolDescription
m9k_searchSearch indexed conversations. Returns compact snippets. Current project boosted. Supports since/until date filters and order (score, date_asc, date_desc).
m9k_contextGet a chunk with surrounding context (adjacent chunks in the same session).
m9k_fullRetrieve full content of chunks by IDs.

Progressive retrieval pattern - search returns ~50 tokens/result, context ~200-300, full ~500-1000. Start with m9k_search, drill down only when needed. 4x token savings vs loading everything.

Context-aware ranking - results from your current project (×1.5) and current session (×1.2) are automatically promoted. Cross-project results remain visible.

Specialized search

ToolDescription
m9k_file_historyFind past conversations that touched a specific file.
m9k_errorsFind past solutions for an error message.
m9k_similar_workFind past approaches to similar tasks. Prioritizes rich metadata.

Memory management

ToolDescription
m9k_saveManually save a memory note for future recall.
m9k_sessionsList all indexed sessions, optionally filtered by project.
m9k_infoShow memory index info: corpus size, search pipeline, embedding worker, usage metrics.
m9k_configView or update plugin configuration.
m9k_forgetPermanently remove a chunk from the index.
m9k_delete_sessionDelete a session from the index.
m9k_ignore_projectExclude a project from indexing. Future sessions won't be indexed, existing ones optionally purged.
m9k_unignore_projectRe-enable indexing for a previously ignored project. Purged data is not restored.
m9k_restartRestart the MCP server to load fresh code after npm run build. Supports force: true for stuck processes.

Usage guide

ToolDescription
__USAGE_GUIDEPhantom tool. Its description teaches Claude the retrieval pattern and available tools.

Configuration

Zero config by default. Everything is tunable via m9k_config or environment variables.

SettingDefaultEnv var
Database path~/.melchizedek/memory.dbM9K_DB_PATH
Daemon modeenabledM9K_NO_DAEMON=1 to disable
Log levelwarnM9K_LOG_LEVEL
Embeddings enabledtrueM9K_EMBEDDINGS=false to disable
Reranker enabledtrueM9K_RERANKER=false to disable

See docs/configuration.md for the full settings reference (20+ options, env vars, config file examples).

Enhanced Search

Melchizedek works out of the box with BM25 keyword search. Text embeddings (MiniLM) download automatically on first use for semantic search.

For GPU-accelerated code embeddings (Ollama), cross-encoder reranking (GGUF models), platform-specific setup guides, and the full model reference, see Enhanced Search Setup.

How is this different?

Melchizedekclaude-historian-mcpclaude-memepisodic-memorymcp-memory-service
[GitHub stars] [npm][GitHub stars] [npm][GitHub stars] [npm][GitHub stars][GitHub stars] [PyPI]
PhilosophySearch engine - indexes everything, you searchSearch engine - scans JSONL on demandNotebook - AI compresses & savesSearch engineNotebook - AI decides what to store
Indexes raw conversationsYes (JSONL transcripts)Yes (direct JSONL read, no persistent index)Compressed summariesYes (JSONL)No (manual store_memory)
Retroactive on installYes (backfills all history)Yes (reads existing files)NoYesNo (empty at start)
SearchBM25 + vectors + RRF + rerankerTF-IDF + fuzzy matchingFTS5 + ChromaDBVectors onlyBM25 + vectors
Progressive retrieval3 layers (search/context/full)NoNoNoNo
100% offlineYesYesNo (needs API for compression)YesYes
Single-file storageSQLiteNone (reads raw JSONL)SQLite + ChromaDBSQLiteSQLite-vec
Zero configYesYesYesYesYes
MCP tools16104212
LicenseMITMITAGPL-3.0MITApache-2.0
Dual embedding (text + code)Yes (MiniLM + Jina Code)NoNoNoNo
Configurable modelsYes (Transformers.js or Ollama)NoNo (Chroma internal)No (hardcoded)Yes (ONNX, Ollama, OpenAI, Cloudflare)
RerankerCross-encoder (ONNX, GGUF, or HTTP)NoNoNoQuality scorer (not search reranker)
PrivacyAll local, <private> tag redactionAll localSends data to Anthropic APIAll localAll local
Multi-instanceSingleton daemon - N Claude windows share 1 process (Unix socket / Windows named pipe, local fallback)N separate processesShared HTTP worker (:37777)N separate processesShared HTTP server

Inspirations

This project stands on the shoulders of others. Key ideas borrowed from:

ProjectWhat we took
CASSRRF hybrid fusion, SHA-256 dedup, auto-fuzzy fallback[GitHub stars]
claude-historian-mcpSpecialized MCP tools (file_history, error_solutions)[GitHub stars] [npm]
claude-diaryPreCompact hook (archive before /compact)[GitHub stars]

Known issues

  • Session boost inactive - Claude Code currently sends an empty session_id in the SessionStart hook stdin payload, preventing the ×1.2 session boost from working. The ×1.5 project boost is unaffected and provides the primary context-aware ranking. Related upstream issues: #13668 (empty transcript_path), #9188 (stale session_id). Melchizedek's session boost code is tested and ready, and will activate automatically when the upstream fix lands.

Privacy

  • Zero telemetry. No tracking, no analytics, no network calls (except optional lazy model download).
  • Read-only on transcripts. Never writes to ~/.claude/projects/. All data in ~/.melchizedek/.
  • <private> tag support. Content between <private>...</private> is replaced with [REDACTED] before indexing.
  • Local-only. Your conversations never leave your machine.

Requirements

  • Node.js >= 20
  • Claude Code >= 2.0
  • macOS, Linux, or Windows

License

MIT

"Without father, without mother, without genealogy, having neither beginning of days nor end of life."

  • Hebrews 7:3

Built by @louis49

来源：README.md，提交 3053612

工具

0
SourceWeft 尚未在沙箱中启动过该软件包，因此其工具未知。

版本历史

1
  1. v1.0.2最新Sep 16, 2026

评分

已满足适用检查项的 50%

该评分衡量此收录条目的完整程度，依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价，也不表示服务器安全或当前可用：可连通性是带日期的检查结果，评分也不包含安装量或用户评价。

  • 近 30 天内可连通25 分

    请保持端点能够响应 MCP 握手；SourceWeft 每 30 天探测一次。

  • 已声明认证方式10 分

    请在注册中心条目中声明凭据请求头，或保持端点能够响应探测，以便判定其认证方式。

  • 已列出工具15 分

    请让服务器在无凭据时响应 tools/list，或发布可在沙箱中无凭据启动的软件包。

  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分