Context Continuity MCP Server

MCP (Model Context Protocol) server that exposes the Context Continuity Engine to AI assistants. Provides 10 tools for querying activity history, running semantic searches, predicting context, traversing temporal graphs, and managing privacy controls.

This server is a read/query layer only — it does not start activity monitoring. The Context Continuity Engine daemon runs separately and populates the data stores that this server reads from.

Architecture

┌─────────────────────┐       stdio (JSON-RPC)       ┌──────────────────────┐
│   AI Assistant       │ ◄──────────────────────────► │  MCP Server (this)   │
│   (Claude, etc.)     │                              │  10 query tools      │
└─────────────────────┘                               └──────────┬───────────┘
                                                                 │ imports
                                                      ┌──────────▼───────────┐
                                                      │  Context Continuity  │
                                                      │  Engine (daemon)     │
                                                      ├──────────────────────┤
                                                      │  SQLite DB           │
                                                      │  Vector Embeddings   │
                                                      │  Temporal Graph      │
                                                      │  Privacy Filter      │
                                                      └──────────────────────┘

The server initializes read-only handles to the engine's four core components:

| Component | What it stores | |-----------|---------------| | ActivityDatabase | Tracked activities in SQLite (window titles, apps, timestamps) | | EmbeddingStore | Vector embeddings for semantic similarity search | | TemporalGraph | Activity relationship graph with temporal decay | | ContextPredictor | ML-based predictions combining all three data sources |

A PrivacyFilter sits in front of everything, enforcing app and directory blacklists.

Tools

Query Tools

| Tool | Description | Key Parameters | |------|-------------|----------------| | context_recent_activities | Get recent tracked activities | hours (default 24), limit (default 50) | | context_search | Semantic search across all activities using embeddings | query (required), limit (default 10) | | context_predict | Predict relevant context for a given activity description | activity_description (required), max_results (default 5) | | context_suggestions | Get actionable suggestions (related files, apps, next actions) | activity_description (required) | | context_related | Traverse the temporal graph to find related activities | activity_id (required), max_depth (default 2) | | context_stats | Get statistics from all engine components | — | | context_list_contexts | List tracked work contexts ordered by last active | limit (default 20) |

Management Tools

| Tool | Description | Key Parameters | |------|-------------|----------------| | context_cleanup | Remove activity data older than N days | days (default 90) | | context_privacy_blacklist | Add/remove privacy blacklist entries | type (app|directory), value, action (add|remove) | | context_create_context | Create or update a named work context | name (required), description, tags |

Prerequisites

Context Continuity Engine — installed and running its daemon to populate data stores
- Default expected location: ~/Documents/PythonScripts/ContextContinuityEngine/
Python 3.10+
PyYAML — for reading engine configuration

The server imports directly from the engine's Python packages (context_engine.*), so the engine must be installed or on the Python path.

Setup

1. Clone

git clone https://github.com/NET-OF-BEING/context-continuity-mcp.git

2. Configure Claude Code

Add to your ~/.mcp.json:

{
  "mcpServers": {
    "context-continuity": {
      "command": "/path/to/your/venv/bin/python3",
      "args": ["context_continuity_server.py"],
      "cwd": "/path/to/context-continuity-mcp"
    }
  }
}

Or use the included launcher script:

{
  "mcpServers": {
    "context-continuity": {
      "command": "/path/to/context-continuity-mcp/run.sh"
    }
  }
}

3. Verify

Once configured, the AI assistant should have access to all 10 context_* tools. Ask it to run context_stats to confirm the engine components are connected.

Configuration

The server reads its data paths from the engine's config file at:

ContextContinuityEngine/config/default_config.yaml

Key config sections used:

storage:
  database_path: "data/activities.db"

vector_db:
  collection_name: "activities"
  model: "all-MiniLM-L6-v2"

graph:
  max_nodes: 10000
  decay_factor: 0.95

prediction:
  prediction_window: 3600
  min_confidence: 0.3

privacy:
  # blacklisted apps and directories

Protocol

Communicates over stdio using JSON-RPC 2.0 per the MCP specification (protocol version 2024-11-05).

Reads newline-delimited JSON from stdin
Writes newline-delimited JSON responses to stdout
Diagnostic messages go to stderr

Example Usage

Once connected to an AI assistant, you can ask natural language questions like:

"What was I working on yesterday?" → context_recent_activities
"Find everything related to the API refactor" → context_search
"What context is relevant to writing unit tests?" → context_predict
"What should I do next on this task?" → context_suggestions
"Show me activities connected to this one" → context_related
"How much data is being tracked?" → context_stats
"Clean up anything older than 60 days" → context_cleanup
"Blacklist my banking app from tracking" → context_privacy_blacklist

Context Continuity Engine — the core engine and daemon that powers this server

License

MIT

MCP Servers