cont3xt

Virtual Infinite Context for Agents and LLMs An MCP server that maintains a continuous rolling context window, always surfacing the most relevant memories for the task at hand — while respecting your token budget.

Current Status

This repository includes an MVP MCP stdio server suitable for integration with the Cline extension for VS Code. It exposes a minimal, SQLite-only toolset over stdio with zero external services required by default.

• Version: 0.1.1
• Transport: MCP stdio
• Default storage: SQLite at ./data/memory.db (configurable; example config uses ./data/novel_memory.db)
• Optional backends: Qdrant / Neo4j (disabled by default in this MVP)
• Tools implemented:
• memory_upsert
• search_memory
• context_pack
• health_ping

Notes:

• Tool names use underscores for compatibility with the Python server decorator. Dotted aliases (e.g., context.pack) can be added later.
• All tools return a standard JSON envelope as text content:

{ "ok": boolean, "data": {...}, "meta": { "duration_ms": number, "backend": "sqlite", "version": "0.1.1" } }

Quickstart

1) Install

Requires Python 3.11+

git clone https://github.com/elevend0g/cont3xt.git
cd cont3xt
pip install -r requirements.txt
pip install -e .

2) Initialize local storage (SQLite)

virtual-context-mcp --init-db

This creates ./data/memory.db if it doesn’t exist (or the path specified by config/env).

3) Run (stdio transport)

virtual-context-mcp

This starts the MCP server over stdio.

Using with MCP Clients

Cline (VS Code)

Add an MCP server named cont3xt that launches the stdio server:

Example configuration shape (adapt for your Cline settings UI/JSON):

{
  "mcpServers": {
    "cont3xt": {
      "command": "virtual-context-mcp",
      "args": ["--config", "configs/novel_writing.yaml"],
      "env": {
        "CONTEXT_MAX_TOKENS": "12000",
        "PYTHONUNBUFFERED": "1"
      }
    }
  }
}

Notes:

• Ensure virtual-context-mcp is in PATH (pip install -e . creates the console script).
• The --config argument is optional; environment variables can override values (see Configuration).

Claude Desktop

{
  "mcpServers": {
    "cont3xt": {
      "command": "virtual-context-mcp",
      "args": ["--config", "configs/novel_writing.yaml"],
      "env": {
        "CONTEXT_MAX_TOKENS": "12000",
        "PYTHONUNBUFFERED": "1"
      }
    }
  }
}

Tools (MVP)

All tools return a JSON envelope as text content.

Envelope:

{
  "ok": true,
  "data": { /* tool-specific payload */ },
  "meta": {
    "duration_ms": 5.23,
    "backend": "sqlite",
    "version": "0.1.1"
  }
}

1) memory_upsert

Upsert memory for a session from either combined content or a user/assistant pair.

Arguments:

• session_id: string (required)
• content: string (optional; direct content)
• user_input: string (optional; used if content not provided)
• assistant_response: string (optional; used if content not provided)

Response (example):

{
  "ok": true,
  "data": {
    "session_id": "sess-1",
    "ids": ["b1c...f"],
    "count": 1
  },
  "meta": {
    "duration_ms": 1.23,
    "backend": "sqlite",
    "version": "0.1.1"
  }
}

2) search_memory

Simple substring search with a recency-biased score over recent SQLite chunks.

Arguments:

• session_id: string (required)
• query: string (required)
• max_results: number (optional, default 10)

Response (example):

{
  "ok": true,
  "data": {
    "query": "emerald eyes",
    "results_count": 2,
    "results": [
      {
        "chunk_id": "b1c...f",
        "score": 1.7,
        "timestamp": "2025-08-09T17:30:00.000000",
        "token_count": 142,
        "preview": "User: ... Assistant: ..."
      }
    ]
  },
  "meta": {
    "duration_ms": 2.45,
    "backend": "sqlite",
    "version": "0.1.1"
  }
}

3) context_pack

Packs a token-budgeted context using recent conversation chunks. Reserves a small buffer and includes the current input if provided.

Arguments:

• session_id: string (required)
• current_input: string (optional, default "")
• budget_tokens: number (optional, defaults to CONTEXT_MAX_TOKENS)

Response (example):

{
  "ok": true,
  "data": {
    "pack": {
      "schema_version": "ctx.v1",
      "budget_tokens": 12000,
      "used_tokens": 2834,
      "sections": [
        {"role": "user", "title": "Current Input", "content": "..."},
        {
          "role": "context",
          "title": "Conversation (2025-08-09T17:30:00.000000)",
          "content": "...",
          "chunk_id": "b1c...f",
          "token_count": 142
        }
      ],
      "provenance": {
        "retriever": "recent-only",
        "stores": {"sqlite": "./data/memory.db"}
      },
      "pack_stats": {
        "num_candidates": 12,
        "kept": 4,
        "dropped": 8,
        "buffer_tokens": 500
      }
    },
    "meta": {
      "input_tokens": 120,
      "output_tokens": 2834,
      "budget": 12000,
      "sources": ["b1c...f", "a9d...1"]
    }
  },
  "meta": {
    "duration_ms": 7.89,
    "backend": "sqlite",
    "version": "0.1.1"
  }
}

Notes:

• If nothing fits within the budget (after a fixed buffer of 500 tokens), the pack includes: "reason": "BUDGET_TOO_SMALL".
• Truncation policy: respect budget strictly; sections are assembled most-recent-first, then presented chronologically for readability.

4) health_ping

Returns basic status and configuration fingerprint.

Arguments: none

Response (example):

{
  "ok": true,
  "data": {
    "server": "cont3xt",
    "version": "0.1.1",
    "datetime": "2025-08-09T17:32:00.000000",
    "config": {
      "sqlite_path": "./data/memory.db",
      "max_tokens": 12000,
      "token_model": "cl100k_base",
      "optional_backends": {
        "qdrant": false,
        "neo4j": false
      }
    }
  },
  "meta": {
    "duration_ms": 0.41,
    "backend": "sqlite",
    "version": "0.1.1"
  }
}

Cline Demo Task (Code Gen Flow)

1) In Cline, run a task like “Improve function X in file Y; add a unit test and make tests pass.” 2) Cline should:

• Call search_memory(session_id, query="project goals")
• Call context_pack(session_id, current_input="<file diff request>", budget_tokens=12000)
• Generate plan → apply small diff → run tests
• Call memory_upsert with (user_input, assistant_response) to persist the session outcome

3) Acceptance:

• From a clean clone, pip install -e ., launch Cline, and complete a guided code edit PR in one go using the above flow.

Configuration

Defaults are provided in code, optionally loaded from a YAML file (e.g., configs/novel_writing.yaml) and overridden by environment variables.

Environment variable overrides (examples):

• CONTEXT_MAX_TOKENS (default 12000)
• CONTEXT_PRESSURE_THRESHOLD (default 0.8)
• CONTEXT_RELIEF_PERCENTAGE (default 0.4)
• CONTEXT_CHUNK_SIZE (default 3200)
• CONTEXT_TOKEN_MODEL (default cl100k_base)

• DATABASE_SQLITE_PATH (default ./data/memory.db)
• DATABASE_QDRANT_URL
• DATABASE_QDRANT_COLLECTION
• DATABASE_NEO4J_URL
• DATABASE_NEO4J_USER
• DATABASE_NEO4J_PASSWORD

For the MVP, only SQLite is used. Qdrant / Neo4j are intentionally not required; future versions can enable them via env/config flags.

Truncation policy:

• A fixed buffer of 500 tokens is reserved.
• Packing order: include current_input (if provided) and then pack recent conversation chunks until the budget is met.
• If no sections can be included, the pack includes "reason": "BUDGET_TOO_SMALL".

Development

Entry points:

• Console script: virtual-context-mcp → virtual_context_mcp.__main__:main
• MCP server class: virtual_context_mcp.server:VirtualContextMCPServer

Local logs: standard output (set PYTHONUNBUFFERED=1 for real-time logs). SQLite file: ./data/memory.db by default.

Run unit/integration tests as needed (note that legacy story-focused modules may not reflect the MVP surface area).

Security

• MVP runs in a safe default mode: only persists to local SQLite.
• No filesystem ingestion or external network writes by default.

Roadmap

• Optional Qdrant/Neo4j integration behind env flags with graceful fallback.
• Aliased/dotted tool names for compatibility (e.g., context.pack).
• Richer retrieval (hybrid, deduplication), entity/graph memory, and pressure relief policies.
• DevContainer / Docker image for reproducible environments.

License

MIT © elevend0g