ultrasync

ultrasync exposes an MCP server aimed at semantic and multi-faceted indexing and search for codebases, enabling coding agents to understand and explore repositories efficiently. It supports semantic vector search, lexical BM25-like keyword matching, and pattern-based memory of insights, decisions, and constraints. Tools exposed via the MCP server cover indexing, searching, memory operations, session context handling, pattern scanning, anchors, and conventions, allowing agents to index files, perform memory recall, run pattern scans, and retrieve contextual information with minimal LLM usage. To use it, install ultrasync via the UV tool, configure the ultrasync MCP entry in your agent’s MCP configuration, and run the server locally. The integration is designed for low friction, with a default toolset focused on search and memory, and optional remote synchronization if needed.

Prerequisites:

• Python and/or UV tool installed (as required by your environment).
• Access to install and run MCP tools (uv CLI for Ultrasync).

Install the Ultrasync MCP tooling:

uv tool install "ultrasync-mcp[cli,lexical,secrets]"

(Optional: include sync features if you have remote synchronization needs)

uv tool install "ultrasync-mcp[cli,lexical,secrets,sync]"

Configure the MCP server entry (as shown in the README):

{
  "ultrasync": {
    "type": "stdio",
    "command": "uv",
    "args": [
      "tool",
      "run",
      "--from",
      "ultrasync-mcp",
      "ultrasync",
      "mcp"
    ],
    "env": {
      "ULTRASYNC_TOOLS": "search,memory"
    }
  }
}

Run and verify:

# Ensure uv is available and the ultrasync-mcp tools are installed
# Start the UDP/stdio MCP integration as configured above (depends on your orchestration)

Follow the agent documentation to point your MCP client at the ultrasync server using the configured entry name (e.g., ultrasync) and the provided type/command/args.

Tips and notes:

• The ultrasync storage uses a dedicated .ultrasync directory for index and blob data; add this to your .gitignore to avoid committing large artifacts).
• Default tools include search and memory; add sync to ULTRASYNC_TOOLS if you plan to use remote synchronization with the provided API.
• Supported MCP capabilities span indexing (index_file, full_index, etc.), multiple search modalities (semantic, lexical, hybrid), memory operations, session threads, pattern scanning, anchors, and conventions. Review and tailor the environment and tool list to your workflow.
• If you encounter performance issues, check the LMDB and vector storage health, and ensure that the underlying filesystem supports the required atomic operations.
• When running in multi-process environments, ensure proper locking and session management to avoid race conditions in memory/index updates.