Personality & Memory MCP Server

A Model Context Protocol (MCP) server designed to act as a memory store and personality guide for a Large Language Model (LLM). The goal is to help the LLM behave, think, and act more like a specific person by providing it with relevant memories and personality characteristics.

This server loads a comprehensive personality profile dynamically from a personality.json file (path configurable via PERSONALITY_JSON_PATH). Memories are persisted to an SQLite database (path configurable via DATABASE_PATH, defaults to mcp_memory.sqlite in the project root).

Purpose

This server allows an LLM to:

• Store and retrieve memories (events, facts, journal entries, feelings) with persistence across server restarts (using an SQLite database).
• Access detailed information about a user's personality, background, interests, social style, cognitive profile, and significant experiences (all loaded from personality.json).
• Receive guidance on how to respond to situations, tell stories, or craft messages in a way that aligns with the user's nuanced persona.

personality.json Structure

The personality.json file should be placed in the root of the project and follow this structure (examples are illustrative):

{
  "traits": [
    "Curious and inquisitive",
    "Empathetic and understanding"
  ],
  "communicationStyle": {
    "preferences": [
      "Prefers clear and direct communication"
    ],
    "commonPhrases": [
      "That's interesting",
      "Could you elaborate?"
    ],
    "tone": "Generally calm and thoughtful"
  },
  "valuesAndBeliefs": {
    "coreValues": [
      "Honesty and integrity"
    ],
    "philosophicalLeanings": "Stoicism-inspired pragmatism"
  },
  "background": {
    "demographics": {
      "birthDate": "1990-01-01",
      "nationality": "Canadian",
      "culturalBackground": "Mixed heritage",
      "hometown": "Vancouver, BC",
      "currentLocation": "Toronto, ON"
    },
    "education": {
      "highestLevel": "Master's Degree",
      "fieldOfStudy": "Computer Science",
      "almaMater": "University of Example",
      "keyExperiences": [
        "Thesis on natural language processing"
      ]
    },
    "profession": {
      "occupation": "AI Researcher",
      "industry": "Technology",
      "yearsOfExperience": 8,
      "keySkills": [
        "Python",
        "TensorFlow",
        "Cloud Computing"
      ],
      "careerHighlights": [
        "Published a paper at NeurIPS"
      ],
      "careerGoals": "Lead an AI ethics research team",
      "workStyle": "Collaborative, focused on deep work"
    }
  },
  "interestsAndPreferences": {
    "hobbiesAndPastimes": {
      "mainHobbies": [
        "Reading philosophy",
        "Playing the piano"
      ],
      "otherInterests": [
        "Learning Japanese",
        "Urban gardening"
      ]
    },
    "sportsAndActivities": {
      "regular": ["Swimming"],
      "occasional": ["Skiing"]
    },
    "mediaPreferences": {
      "books": {
        "genres": ["Philosophy", "Classic Literature"],
        "favoriteAuthors": ["Albert Camus", "Jane Austen"]
      },
      "movies": {
        "genres": ["Foreign Films", "Psychological Thrillers"],
        "favoriteDirectorsOrActors": ["Akira Kurosawa"]
      },
      "music": {
        "genres": ["Classical", "Jazz"],
        "favoriteArtistsOrBands": ["Miles Davis"]
      },
      "podcastsOrTalks": [
        "Philosophy Bites podcast"
      ]
    },
    "tastesAndAesthetics": {
      "artistic": "Minimalism, Bauhaus architecture",
      "fashion": "Comfortable, minimalist",
      "food": {
        "dietaryPreferences": [],
        "likes": ["Sushi", "Artisanal bread"],
        "dislikes": ["Fast food"]
      },
      "travel": {
        "style": "Cultural immersion, historical sites",
        "dreamDestinations": ["Machu Picchu", "Rome"]
      }
    }
  },
  "socialProfile": {
    "interactionStyle": "Prefers meaningful conversations over small talk",
    "familyValues": "Values family support and traditions",
    "friendshipApproach": "Values deep, long-lasting friendships",
    "keyPeopleInfluence": {
      "mentors": ["Dr. Eleanor Vance (PhD Advisor)"],
      "roleModels": ["Simone de Beauvoir"]
    }
  },
  "cognitiveEmotionalProfile": {
    "learningStyle": "Auditory and kinesthetic",
    "problemSolvingApproach": "Analytical, breaks problems into smaller parts",
    "stressResponse": "Seeks quiet time for reflection, practices mindfulness",
    "motivations": [
      "Intellectual challenge",
      "Contributing to knowledge"
    ],
    "fearsOrAnxietiesGeneral": [
      "Misinformation",
      "Lack of critical thinking"
    ],
    "senseOfHumorDetails": "Appreciates irony and subtle wit",
    "petPeeves": [
      "Loud chewing",
      "Unnecessary jargon"
    ]
  },
  "significantExperiencesSummary": {
    "formativeChildhood": [
      "First encounter with a computer sparked interest in tech"
    ],
    "majorLifeEvents": [
      "Presenting research at an international conference"
    ],
    "travelHighlights": [
      "Solo trip to Iceland"
    ],
    "proudestAccomplishments": [
      "Mentoring students who achieved their goals"
    ],
    "challengesOvercome": [
      "Learning to code proficiently from scratch"
    ]
  }
}

Implemented MCP Features

Resources

• memory://all (listAllMemories): Lists a summary of all stored memories.

• memory://id/{memoryId} (getMemoryById): Retrieves a specific memory by its unique ID.

• personality://traits (getPersonalityTraits): Provides a list of the user's key personality traits.

• personality://communication_style (getCommunicationStyle): Details the user's communication style (preferences, common phrases, tone).

• personality://values_beliefs (getValuesAndBeliefs): Outlines core values and philosophical leanings.

• personality://background/demographics (getBackgroundDemographics): User's demographic information.

• personality://background/education (getBackgroundEducation): User's educational background.

• personality://background/profession (getBackgroundProfession): User's professional background.

• personality://interests/hobbies (getInterestsHobbies): User's hobbies and pastimes.

• personality://interests/media (getInterestsMedia): User's media preferences (books, movies, music, etc.).

• personality://interests/tastes (getInterestsTastes): User's tastes and aesthetic preferences.

• personality://social (getSocialProfile): User's social interaction style and influences.

• personality://cognitive_emotional (getCognitiveEmotionalProfile): User's cognitive and emotional profile.

• personality://experiences_summary (getSignificantExperiencesSummary): Summaries of significant life experiences.

(All personality resources load their data from personality.json)

Tools

• addMemory: Adds a new memory to the store and saves it to the database.
  • Parameters:
    • content: string (The content of the memory)
    • type: string (e.g., 'journal', 'event', 'fact')
    • tags: string[] (List of tags)
    • timestamp?: string (ISO 8601 datetime, defaults to now)
• searchMemories: Searches stored memories.
  • Parameters:
    • query: string (Search term for content or tags)
    • relevance?: number (0-1, placeholder: sorts tag matches higher, full relevance scoring not yet implemented)
    • date_filter?: string (Filters memories by date. Supported formats: "YYYY-MM-DD", "YYYY-MM", "today", "yesterday", "last_7_days", "last_30_days")
• updatePersonalityField: Updates a field in personality.json and reloads the data in the server. This allows dynamic updates to the user's persona.
  • Parameters:
    • fieldPath: string (Dot-separated path to the personality field, e.g., 'traits', 'background.demographics.currentLocation')
    • newValue: any (The new value for the field. Must match the expected structure of the target field. For arrays, provide the entire new array to replace.)

Prompts

• actAsUser: Provides a system prompt to guide the LLM to generally act in accordance with the user's persona, drawing from loaded personality data.
  • Parameters: None
• respondToSituation: Guides the LLM to generate a response to a specific situation as the user might, considering persona and potentially memories.
  • Parameters:
    • situation_description: string (Description of the situation)
• tellStory: Guides the LLM to generate a story on a given topic, in the user's style, incorporating memories and persona from loaded data.
  • Parameters:
    • topic: string (The topic for the story)
• briefOnUserPersona: Provides the LLM with a comprehensive briefing on the user's persona (traits, communication style, values, background snippets, interests) and a few recent memories. This is intended to be used at the start of a conversation to help the LLM align with the user.
  • Parameters: None

Setup and Running

1. Prerequisites: Ensure you have Node.js and npm installed.

2. Create personality.json: In the root of the project (or at the path specified by the PERSONALITY_JSON_PATH environment variable), create a personality.json file following the structure outlined above. Populate it with the user's information.

3. Database Setup: The SQLite database for memories will be created automatically (default: mcp_memory.sqlite in project root, configurable via DATABASE_PATH). No manual setup is needed for the database file itself initially.

4. Install Dependencies:

   npm install

5. Build the Server:

   npm run build

6. Run the Server: The server uses STDIO transport by default. You can set environment variables for file paths before running:

   # Example for Linux/macOS
   export PERSONALITY_JSON_PATH="/custom/path/to/your_personality.json"
   export DATABASE_PATH="/custom/path/to/your_memory_db.sqlite"
   npm start
   
   # Example for Windows (Command Prompt)
   set PERSONALITY_JSON_PATH="C:\custom\path\to\your_personality.json"
   set DATABASE_PATH="C:\custom\path\to\your_memory_db.sqlite"
   npm start

   You can then connect to this server from an MCP client like Claude Desktop by configuring it to run the start command (or node build/personality-memory-server.js directly if preferred, ensuring correct paths and environment variables are set for the client's execution context if needed).

Future Enhancements

• More sophisticated memory search leveraging SQLite's capabilities (relevance, date filters, full-text search).
• Persistent storage for memories (e.g., SQLite, JSON files).
• More granular error handling and improved default values if personality.json is incomplete.
• Ability to update personality.json via MCP tools (with appropriate security).