style-memory-mcp

English | 简体中文

A tiny local MCP server that learns a user's conversational style, catchphrases, dialect markers, emoji habits, tone preferences, and concrete collaboration preferences without storing private memories.

It is meant to remember the flavor of how someone talks, not the private facts of their life.

Why

Most agent memory tools remember facts:

• "The user lives in..."

• "The user works on..."

• "The user prefers..."

style-memory-mcp remembers voice:

• "The user says lol or 哈哈哈 casually."

• "The user likes warm, playful replies."

• "The user sometimes uses Sichuan markers like 锤子 or 巴适."

• "The user uses kaomoji like (｡･ω･｡)."

Small thing. Big vibe. (｡･ω･｡)ﾉ

Related MCP server: my-voice-mcp

Features

• Local JSON store only — your data stays on your machine

• No cloud service, no telemetry, no external API calls

• The MCP server itself never calls an LLM. The dictionary path is pure regex. You may optionally let the host agent forward its own observations via hints — see LLM-assisted learning below.

• No full conversation log storage — only style signals (and a short ≤60-char usage example per habit, sanitized before storage)

• Learns candidates first, promotes repeated habits later

• Promotion now also requires the habit to appear under ≥2 distinct context labels (cross-context check, inspired by nuwa-skill)

• Auto-cleans stale habits (candidate → archived → deleted)

• Supports Chinese, English, emoji, kaomoji, and dialect markers — plus free-form idiolect for whatever the host LLM notices

• Built-in dictionary covers Sichuan, Cantonese, Northeast (Dongbei), Shanghainese, and Min Nan / Taiwanese markers, plus current (2024–2026) Chinese and English internet slang. Locale-tagged so the agent can tell universally-safe phrases apart from slang that must stay out of legal / medical / serious replies.

• Returns an actionable style brief: how to apply the style first, then the context-relevant habits

• Supports an interaction profile: how the user prefers the agent to collaborate, without personality labels

• Interaction-profile preferences can be reviewed, pinned, or forgotten just like style habits

• Includes a lightweight get_style_memory_score health check for readiness, drift risk, over-imitation risk, and brief refresh recommendations

• Works with any MCP-capable agent that calls the tools

• Pin habits to protect them from auto-cleanup

• Pause learning anytime with set_learning_enabled

Installation

Local install

git clone https://github.com/hexingyuofficial/style-memory-mcp.git
cd style-memory-mcp
npm install
npm run build

Global install (optional)

npm install -g style-memory-mcp
# Then use: style-memory-mcp

For local development:

npm run dev

MCP Client Config

Add to your MCP client configuration (e.g. Claude Desktop, Cursor, etc.):

{
  "mcpServers": {
    "style-memory": {
      "command": "node",
      "args": ["/absolute/path/to/style-memory-mcp/dist/server.js"]
    }
  }
}

You can customize the JSON store location:

{
  "mcpServers": {
    "style-memory": {
      "command": "node",
      "args": ["/absolute/path/to/style-memory-mcp/dist/server.js"],
      "env": {
        "STYLE_MEMORY_PATH": "/absolute/path/to/style-memory.json"
      }
    }
  }
}

See docs/INTEGRATION.zh-CN.md for a practical Chinese integration guide, including Doubao-style setup notes and the recommended automatic brief refresh protocol.

Environment Variables

Tools

observe_user_message

Learns lightweight style signals from the latest user message.

Agents should call this after user messages, but should not send secrets, private memory dumps, or full chat logs.

May optionally include a hints array — see LLM-assisted learning below — and a profileHints array for concrete collaboration preferences.

get_style_brief

Returns a short style brief for the agent to use lightly.

Agents should call this at the start of a conversation, or before drafting a friendly reply.

distill_recent_style

Batched, user-endorsed seeding. The host LLM submits 3–8 high-conviction observations distilled from recent messages. Each habit becomes active immediately. Useful for warm-starting a fresh store, or when the user explicitly asks the agent to "really learn how I talk".

distill_interaction_profile

Batched, user-endorsed seeding for concrete collaboration preferences such as "prefers value judgment before steps" or "likes plan → implement → verify for technical work". Do not submit personality labels, psychological states, diagnoses, or private facts.

list_style_habits

Lists candidates, active habits, and archived habits.

list_interaction_profile

Lists stored collaboration and response-structure preferences.

review_style_habits

Returns a short review queue with suggested actions: keep, pin, forget, or observe. Useful when the user wants to inspect what the MCP has learned.

review_interaction_profile

Returns a short review queue for collaboration preferences with suggested actions: keep, pin, forget, or observe.

forget_style_habit

Deletes a habit by id or exact text.

forget_interaction_preference

Deletes a collaboration preference by id or exact text.

pin_style_habit

Pins a habit so cleanup will not delete it.

pin_interaction_preference

Pins a collaboration preference so cleanup will not delete it.

set_learning_enabled

Turns learning on or off.

get_style_memory_score

Scores whether the local style memory is usable and stable. Returns readiness, stability, freshness, drift risk, over-imitation risk, whether get_style_brief should be refreshed, and short recommendations.

get_style_memory_status

Shows the JSON path and habit counts.

Agent Instruction

Add something like this to your agent or skill:

Use style-memory-mcp for lightweight conversational style only.
At the start of a conversation, call get_style_brief.
After each user message, call observe_user_message with only the latest user message.
In long chats, silently call get_style_brief again every 12-20 user turns,
after major context switches, before long important answers, or whenever the
user says things like "感觉飘了" or "重新对齐一下".
If you spot a personal habit the built-in dictionary likely wouldn't catch
(e.g. a self-invented sentence-final particle, an unusual structural quirk),
add it as a hints[] entry on the same observe_user_message call.
Three repetitions across two distinct contexts are needed before a habit
becomes stable, so you don't need to be right on the first try.
Do not send secrets, private memories, files, or full conversation logs.
Use returned style hints lightly. Shape the assistant's own stable
collaboration style; never copy the user mechanically.

A longer template lives at examples/agent-instruction.md.

Interaction Profile

style-memory-mcp does not build a personality profile. It can learn concrete, behavioral collaboration preferences that are safer and more useful:

• "The user prefers conclusions before details."

• "For technical work, the user likes plan → implement → verify."

• "The user prefers value judgment before step-by-step instructions."

• "The user dislikes vague praise and wants specific recommendations."

Do not store:

• "The user is anxious."

• "The user is introverted."

• Psychological labels, diagnoses, or personality types.

• Real-world identity, address, job, or other private facts.

Host agents can submit profileHints on observe_user_message:

{
  "text": "First tell me whether this is worth doing, then give steps.",
  "context": "planning",
  "profileHints": [
    {
      "category": "response_structure",
      "text": "prefers value judgment before step-by-step implementation",
      "example": "First tell me whether this is worth doing, then give steps.",
      "useWhen": ["planning", "technical_chat"],
      "confidence": 0.7
    }
  ]
}

For a one-shot seed, use distill_interaction_profile with 1–8 high-conviction preferences. Active profile preferences appear in get_style_brief alongside style habits, but the brief stays short and context-filtered.

If a profile preference is wrong, use forget_interaction_preference. If it is important and should survive cleanup, use pin_interaction_preference. Use review_interaction_profile for a short correction queue.

Drift and Refresh

The MCP server cannot push context into the host agent by itself. The host agent should refresh its alignment brief:

• at the start of a new chat,

• every 12–20 user turns in long chats,

• after major topic or context switches,

• before long or important answers,

• when the user says "感觉飘了", "重新对齐一下", "不像我", or similar.

For a quick health check, call get_style_memory_score. If briefRefreshRecommended is true, call get_style_brief before the next substantial reply.

Read-only Reuse and Restarts

MCP processes are normally started and restarted by the host agent. style-memory-mcp does not need to self-restart. The durable part is the JSON store: if multiple conversations use the same STYLE_MEMORY_PATH, they read the same style memory after any restart.

If the store has learned enough and you want it to guide style without continuing to learn, use this pattern:

1. Keep the same STYLE_MEMORY_PATH.

2. Call get_style_brief at the start of a new conversation.

3. Call set_learning_enabled(false) or set STYLE_MEMORY_LEARNING=off.

4. Turn learning back on only when you want to refresh the style.

This gives you persistent style carryover without writing new habits on every message.

LLM-assisted learning

The dictionary path knows only what's hard-coded (Sichuan dialect, common Chinese/English catchphrases, kaomoji, etc.). It will miss anything the author didn't think of — including the personal habits that make someone sound like themselves.

style-memory-mcp solves this without taking on an LLM dependency itself: the host agent already reads every user message to generate its reply, so let it pass along anything it noticed. The MCP server stays a thin "counter + lifecycle + safety" layer over local JSON. No API key. No network. No model registry. Zero added cost.

// observe_user_message input
{
  "text": "今天天气好巴适莫",
  "context": "casual_chat",
  "hints": [
    {
      "kind": "sentence_final_particle",
      "text": "莫",
      "example": "今天天气好巴适莫",
      "confidence": 0.6
    }
  ]
}

After three observations across two distinct context labels, 莫 is promoted to active and shows up in future get_style_brief calls, example included. High-confidence hints (≥ ~0.71) skip the cross-context gate.

For batched, user-endorsed seeding, call distill_recent_style once with 3–8 observations distilled from recent messages.

Guardrails that make this safe:

• The MCP server itself never calls an LLM — it just records what the host reported. "No network" is still true.

• Hints with a bad kind or empty text are dropped, not learned.

• Examples are sanitized (sanitizeExample): whitespace collapse, length cap, sensitive content (credentials/tokens) silently dropped.

• The three-strike + cross-context promote rule keeps a single hallucinated hint from polluting the active habit set.

• All existing controls (forget_style_habit, pin_style_habit, set_learning_enabled) work unchanged.

Cleanup Rules

The server does not need a background daemon. Cleanup happens when the MCP starts and when tools are called.

Default behavior:

• Candidate habits disappear after 30 inactive days.

• Active habits are archived after 180 inactive days.

• Archived habits are deleted after another 180 inactive days.

• Pinned habits are never deleted automatically.

Important: a habit is refreshed only when the user says it again. Agent usage does not keep it alive, so the system does not get stuck imitating itself.

Example JSON

{
  "id": "zh-cn-sichuan-dialect_marker-锤子",
  "kind": "dialect_marker",
  "text": "锤子",
  "locale": "zh-CN-sichuan",
  "confidence": 0.64,
  "seenCount": 4,
  "status": "active",
  "pinned": false,
  "useWhen": ["casual_chat", "joking", "warm_chat"],
  "avoidWhen": ["serious_debugging", "legal", "medical", "user_upset"]
}

Development

# Install dependencies
npm install

# Type-check
npm run check

# Build
npm run build

# Run tests
npm test

# Development mode (auto-reload with tsx)
npm run dev

Dictionary size & token cost

The built-in dictionary (dialect markers, catchphrases, internet slang) lives in src/extract.ts and is never sent to the LLM. It only participates in local text.includes() / regex scans. Doubling the dictionary costs zero extra tokens per turn.

The only payloads that reach the host LLM are:

1. get_style_brief output — bounded by STYLE_MEMORY_MAX_BRIEF_ITEMS (default 8). The brief surfaces only habits the user has actually exhibited and which were promoted to active, not whatever sits in the dictionary.

2. Tool descriptions — fixed in server.ts, independent of dictionary size.

So if your dialect or slang isn't covered, please send a PR with new entries — it only improves recall and won't bloat anyone's prompts.

Privacy

This project is intentionally boring about data:

• It stores style signals, not raw messages.

• It avoids learning from obvious secret contexts (credential-like patterns are filtered).

• It keeps its own JSON store, separate from any user memory database.

• Users can list, forget, pin, or disable learning at any time.

• No network calls. Everything runs locally.

Contributing

Contributions are welcome! Especially:

• New dialect markers (Cantonese, Shanghainese, Dongbei, etc.)

• New catchphrase patterns for any language

• Better heuristics for sensitivity detection

• Performance improvements

Please add tests for new extraction rules. See src/extract.test.ts for examples.

License

MIT