Content
<img align="right" src="claude-historian.svg" alt="claude-historian-mcp" width="220">
# claude-historian-mcp
An [Model Context Protocol (MCP)](https://modelcontextprotocol.io/) server for searching your [Claude Code](https://docs.anthropic.com/en/docs/claude-code) conversation history. Find past solutions, track file changes, and learn from previous work.
<br clear="right">
[](https://www.npmjs.com/package/claude-historian-mcp) [](https://opensource.org/licenses/MIT) [](https://www.typescriptlang.org/) [](https://nodejs.org/) [](#) [](https://github.com/Vvkmnn/claude-historian-mcp) 
---
## install
**Requirements:**
[](https://claude.ai/code)
**From shell:**
```bash
claude mcp add claude-historian-mcp -- npx claude-historian-mcp
```
**From inside Claude** (restart required):
```
Add this to our global mcp config: npx claude-historian-mcp
Install this mcp: https://github.com/Vvkmnn/claude-historian-mcp
```
**From any manually configurable `mcp.json`**: (Cursor, Windsurf, etc.)
```json
{
"mcpServers": {
"claude-historian-mcp": {
"command": "npx",
"args": ["claude-historian-mcp"],
"env": {}
}
}
}
```
There is **no `npm install` required** -- no external dependencies or local databases, only search algorithms.
However, if `npx` resolves the wrong package, you can force resolution with:
```bash
npm install -g claude-historian-mcp
```
> **renamed:** This project was renamed from `claude-historian` to `claude-historian-mcp`. Existing users should update your install command and MCP config args to `claude-historian-mcp`.
## [skill](.claude/skills/claude-historian)
Optionally, install the skill to teach Claude when to proactively use historian:
```bash
npx skills add Vvkmnn/claude-historian-mcp --skill claude-historian --global
# Optional: add --yes to skip interactive prompt and install to all agents
```
This makes Claude automatically check your history before web searches, when encountering errors, or at session start. The MCP works without the skill, but the skill improves discoverability.
## [plugin](https://github.com/Vvkmnn/claude-emporium)
For automatic history search with hooks and commands, install from the [claude-emporium](https://github.com/Vvkmnn/claude-emporium) marketplace:
```bash
/plugin marketplace add Vvkmnn/claude-emporium
/plugin install claude-historian@claude-emporium
```
The **claude-historian** plugin provides:
**Hooks** (targeted, zero overhead on success):
- Before WebSearch/WebFetch → Check `search scope="similar"`
- Before EnterPlanMode → Check `search scope="plans"`
- Before Task agents → Check `search scope="tools"`
- After Bash errors → Check `search scope="errors"`
**Command:** `/historian-search <query>`
Requires the MCP server installed first. See the emporium for other Claude Code plugins and MCPs.
## features
[MCP server](https://modelcontextprotocol.io/) that gives Claude access to your conversation history. Two tools, 11 scopes, zero dependencies.
Runs locally (with cool shades `[⌐■_■] 📜`):
#### `search`
Search across conversations, files, errors, plans, config, tasks, sessions, tools, similar queries, and memories.
```
search query="docker auth error" # default scope: all
search query="fix build" scope="conversations" # past solutions
search query="ENOENT" scope="errors" # error patterns + fixes
search query="auth" scope="plans" # implementation plans
search query="hooks" scope="config" # rules, skills, CLAUDE.md
search query="git push" scope="similar" # related questions asked before
search query="Edit" scope="tools" # tool usage workflows
search filepath="package.json" scope="files" # file change history
search scope="sessions" # recent sessions
search scope="memories" # project memory files
search query="deploy" scope="all" detail_level="detailed" # full context
search query="auth" timeframe="7d" project="my-app" # filtered
```
```json
📜 ── search "docker auth" ── 5 results · 405 tokens
{
"results": [{
"type": "assistant",
"ts": "2h ago",
"content": "Fixed Docker auth by updating registry credentials...",
"project": "my-app",
"score": 100,
"ctx": { "filesReferenced": ["docker-compose.yml"], "toolsUsed": ["Edit", "Bash"] }
}]
}
```
```json
📜 ── files "package.json" ── 92 operations · 1594 tokens
{
"filepath": "package.json",
"operations": [{
"type": "edit",
"ts": "1d ago",
"changes": ["Changed: \"version\": \"1.0.3\" → \"version\": \"1.0.4\""],
"content": "Updated version for release"
}]
}
```
```json
📜 ── tools "Bash" ── 5 patterns · 427 tokens
{
"tool": "Bash",
"patterns": [{
"name": "Bash",
"uses": 10,
"workflow": "$ npm run build 2>&1",
"practice": "Used with: ts, js, json, md files"
}]
}
```
#### `inspect`
Get an intelligent summary of any session by ID (full UUID or short prefix).
```
inspect session_id="latest" # most recent session
inspect session_id="d537af65" # short prefix works
inspect session_id="d537af65" focus="files" # only file changes
inspect session_id="d537af65" focus="tools" # only tool usage
inspect session_id="d537af65" focus="solutions" # only solutions
```
```json
📜 ── inspect my-app (68d5323b)
{
"session": {
"id": "68d5323b",
"ts": "2h ago",
"duration": 45,
"messages": 128,
"project": "my-app",
"tools": ["Edit", "Bash", "Read"],
"files": ["src/auth.ts", "package.json"],
"accomplishments": ["fixed auth bug", "added unit tests"],
"decisions": ["chose JWT over sessions"]
}
}
```
## methodology
How [claude-historian-mcp](https://github.com/Vvkmnn/claude-historian-mcp) [works](https://github.com/Vvkmnn/claude-historian-mcp/tree/master/src):
```
"docker auth" query
|
├─> Parallel Processing (search.ts:174): 15 projects × 10 files concurrently
| • Promise.allSettled for 6x speed improvement
| • Early termination when sufficient results found
| • Enhanced file coverage with comprehensive patterns
|
├─> Enhanced Classification (search.ts:642): implementation → boost tool workflows
| • Workflow detection for tool sequences (Edit → Read → Bash)
| • Semantic boundary preservation (never truncate mid-function)
| • Claude-optimized formatting with rich metadata
|
├─> Smart Ranking (utils.ts:267):
| ├─> Core Terms (scoring-constants.ts): "docker" +10, "auth" +10
| ├─> Supporting Terms: context words +3 each
| ├─> Tool Usage: Edit/Bash references +5
| ├─> File References: paths/extensions +3
| └─> Project Match: current project +5
|
├─> Results sorted by composite score:
| • "Edit workflow (7x successful)" (2h ago) ***** [score: 45]
| • "Docker auth with context paths" (yesterday) **** [score: 38]
| • "Container debugging patterns" (last week) *** [score: 22]
|
└─> Return Claude Code optimized results
```
**Core optimizations:**
- [parallel processing](https://github.com/Vvkmnn/claude-historian-mcp/blob/master/src/search.ts#L174): `Promise.allSettled` for 6x speed improvement across projects and files
- [workflow detection](https://github.com/Vvkmnn/claude-historian-mcp/blob/master/src/search.ts#L1122): Captures tool sequences like "Edit → Read → Bash" patterns
- [enhanced file matching](https://github.com/Vvkmnn/claude-historian-mcp/blob/master/src/search.ts#L793): Comprehensive path variations with case-insensitive matching
- [intelligent deduplication](https://github.com/Vvkmnn/claude-historian-mcp/blob/master/src/search-helpers.ts#L40): Content-based deduplication preserving highest-scoring results
- [intelligent truncation](https://github.com/Vvkmnn/claude-historian-mcp/blob/master/src/formatter.ts#L46): Never truncates mid-function or mid-error
- [Claude-optimized formatting](https://github.com/Vvkmnn/claude-historian-mcp/blob/master/src/formatter.ts#L26): Rich metadata with technical content prioritization
**Search strategies:**
- **[JSON streaming parser](https://en.wikipedia.org/wiki/JSON_streaming)** ([parseJsonlFile](https://github.com/Vvkmnn/claude-historian-mcp/blob/master/src/parser.ts#L16)): Reads Claude Code conversation files on-demand without full deserialization
- **[LRU caching](<https://en.wikipedia.org/wiki/Cache_replacement_policies#Least_recently_used_(LRU)>)** ([messageCache](https://github.com/Vvkmnn/claude-historian-mcp/blob/master/src/search.ts#L27)): In-memory cache with intelligent eviction for frequently accessed conversations
- **[TF-IDF inspired scoring](https://en.wikipedia.org/wiki/Tf%E2%80%93idf)** ([calculateRelevanceScore](https://github.com/Vvkmnn/claude-historian-mcp/blob/master/src/utils.ts#L267)): Term frequency scoring with document frequency weighting for relevance
- **[Query classification](https://en.wikipedia.org/wiki/Text_classification)** ([classifyQueryType](https://github.com/Vvkmnn/claude-historian-mcp/blob/master/src/search.ts#L642)): Naive Bayes-style classification (error/implementation/analysis/general) with adaptive limits
- **[Edit distance](https://en.wikipedia.org/wiki/Edit_distance)** ([calculateQuerySimilarity](https://github.com/Vvkmnn/claude-historian-mcp/blob/master/src/search-helpers.ts#L157)): Fuzzy matching for technical terms and typo tolerance
- **[Exponential time decay](https://en.wikipedia.org/wiki/Exponential_decay)** (getTimeRangeFilter): Recent messages weighted higher with configurable half-life
- **[Parallel file processing](https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Promise/allSettled)** ([getErrorSolutions](https://github.com/Vvkmnn/claude-historian-mcp/blob/master/src/search.ts#L985)): Concurrent project scanning with early termination for 0.8s response times
- **[Workflow pattern recognition](https://en.wikipedia.org/wiki/Sequential_pattern_mining)** ([getToolPatterns](https://github.com/Vvkmnn/claude-historian-mcp/blob/master/src/search.ts#L1122)): Detects tool usage sequences and related workflows for learning
- **[Enhanced file context](<https://en.wikipedia.org/wiki/Path_(computing)>)** ([findFileContext](https://github.com/Vvkmnn/claude-historian-mcp/blob/master/src/search.ts#L793)): Multi-project search with comprehensive path matching
- **[Content-aware truncation](https://en.wikipedia.org/wiki/Text_segmentation)** ([smartTruncation](https://github.com/Vvkmnn/claude-historian-mcp/blob/master/src/formatter.ts#L46)): Intelligent content boundaries over arbitrary character limits
- **[Technical content prioritization](https://en.wikipedia.org/wiki/Information_extraction)** ([BeautifulFormatter](https://github.com/Vvkmnn/claude-historian-mcp/blob/master/src/formatter.ts#L26)): Code blocks, errors, and file paths get full preservation
- **[Query similarity clustering](https://en.wikipedia.org/wiki/Cluster_analysis)** ([findSimilarQueries](https://github.com/Vvkmnn/claude-historian-mcp/blob/master/src/search.ts#L912)): Semantic expansion and pattern grouping for related questions
**Design principles:**
- **Universal engine** -- single search backend for all Claude Code conversations
- **Parallel processing** -- concurrent file scanning across session directories
- **Semantic expansion** -- query synonyms and related terms for better recall
- **Zero dependencies** -- only `@modelcontextprotocol/sdk`, no databases required
- **Offline** -- never leaves your machine, scans local JSONL files only
**File access:**
- Reads from: `~/.claude/conversations/`
- Zero persistent storage or indexing
- Never leaves your machine
**Performance:** See [PERFORMANCE.md](./PERFORMANCE.md) for benchmarks, optimization history, and quality scores.
## alternatives
Every conversation history tool either loads context always (burning tokens when unused) or requires external runtimes and databases. Historian searches on-demand with zero dependencies.
| Feature | **historian** | Claude Memory | claude-mem | deja | conversation-search |
| ----------------------- | ----------------------- | ----------------------- | -------------------------------- | ------------------- | ---------------------------- |
| **Dependencies** | **Zero** | Built-in | Bun + Python + SQLite + Chroma | Python | Rust toolchain |
| **Background service** | **No** | No | Yes (port 37777) | No | No |
| **Writes to disk** | **Never** | Yes (auto-memory files) | Yes (SQLite + Chroma DB) | Yes (breadcrumbs) | Yes (~10% index overhead) |
| **Session startup** | **0 tokens** | ~200 lines loaded | 5-8k tokens every session | Skill prompt loaded | 0 tokens |
| **Token cost (idle)** | **0** | 200 lines/session | 5-8k/session | Skill prompt/session| 0 |
| **Search algorithms** | **[12](#methodology)** | None (file read) | Vector + keyword | Weighted signals | BM25 full-text |
| **Fuzzy matching** | **Yes** | No | Yes (vector similarity) | No | No |
| **Workflow detection** | **Yes** | No | No | No | No |
| **Raw conversations** | **Yes** | No (summaries only) | No (compressed observations) | Yes | Yes (filtered) |
| **Maintenance** | **Zero** | Zero | Worker daemons, migrations | Skill config | Index rebuilds |
**[Claude Memory](https://docs.anthropic.com/en/docs/claude-code/memory)** -- Claude's built-in memory (`CLAUDE.md` + auto-memory). Persists project rules and preferences across sessions. Forward-looking ("always use ESM imports"); not conversation search. **Complementary**: memory for rules, historian for past solutions.
**[claude-mem](https://github.com/thedotmack/claude-mem)** -- Plugin that captures observations via lifecycle hooks, compresses them into SQLite + Chroma, and loads context every session. Requires Bun, Python, and a background worker on port 37777. Real-world testing (270+ sessions): 95% of sessions never query history -- always-on tools pay 5-8k tokens per session regardless. Historian pays 0 tokens idle, 500-2k per query, saving ~475k tokens over 100 sessions. Known issues: creates stub session files that break `--continue`, worker daemon version conflicts, security hooks blocking valid edits.
**[deja](https://github.com/kateleext/deja)** -- Python skill that indexes sessions by episodes and accomplishments. Uses weighted signal ranking (todos > files > text). Requires Python and TodoWrite integration.
**[conversation-search](https://github.com/ticpu/claude-conversation-search-mcp)** -- Rust MCP server using Tantivy BM25 full-text search. Fast indexing (~1000 conversations/second) but requires Rust toolchain and persistent disk index.
## desktop
**Note:** Claude Desktop stores conversations server-side, not locally. The local LevelDB files (`~/Library/Application Support/Claude/`) contain only session tokens, UI preferences, and Intercom state - not conversation content. Claude Desktop support is also blocked by [LevelDB locks](https://github.com/Level/level#open) and [Electron sandboxing](https://www.electronjs.org/docs/latest/tutorial/sandbox).
This means **local history search for Claude Desktop is not currently possible**. This project focuses on **Claude Code**, which stores full conversation history locally in `~/.claude/projects/`.
You may get some Claude Desktop from Claude Code, but **only when the Claude app is closed**. Furthermore A DXT package and build is available for future compatibility; further investigations are ongoing. Feel free to test with it.
## development
```bash
git clone https://github.com/Vvkmnn/claude-historian-mcp && cd claude-historian-mcp
npm install && npm run build
npm test
```
**Package requirements:**
- **Node.js**: >=20.0.0 (ES modules)
- **Runtime**: `@modelcontextprotocol/sdk`
- **Zero external databases** -- works with `npx`
**Development workflow:**
```bash
npm run build # TypeScript compilation with executable permissions
npm run dev # Watch mode with tsc --watch
npm run start # Run the MCP server directly
npm run lint # ESLint code quality checks
npm run lint:fix # Auto-fix linting issues
npm run format # Prettier formatting (src/)
npm run format:check # Check formatting without changes
npm run typecheck # TypeScript validation without emit
npm run test # Lint + type check
npm run prepublishOnly # Pre-publish validation (build + lint + format:check)
```
**Git hooks (via Husky):**
- **pre-commit**: Auto-formats staged `.ts` files with Prettier and ESLint
- **pre-push**: Runs full validation (format, lint, type-check, build) before push
Contributing:
- Fork the repository and create feature branches
- Test with large conversation histories before submitting PRs
- Follow TypeScript strict mode and [MCP protocol](https://modelcontextprotocol.io/specification) standards
Learn from examples:
- [Official MCP servers](https://github.com/modelcontextprotocol/servers) for reference implementations
- [TypeScript SDK](https://github.com/modelcontextprotocol/typescript-sdk) for best practices
- [Creating Node.js modules](https://docs.npmjs.com/creating-node-js-modules) for npm package development
## license
[MIT](LICENSE)
<hr>
<a href="https://en.wikipedia.org/wiki/Cesare_Maccari"><img src="logo/appius-claudius.jpg" alt="Appius Claudius Caecus in the Senate -- Cesare Maccari" width="100%"></a>
<p align="center">
_**[Appius Claudius Caecus in the Senate](https://en.wikipedia.org/wiki/Cesare_Maccari)** by **[Cesare Maccari](https://en.wikipedia.org/wiki/Cesare_Maccari)** (1888). Roman statesman and father of Latin prose._
</p>
