inked

Inked is a fast memory search service designed to work with an MCP server configuration. It provides a quick, text-based search of stored memories, with an optional AI-powered embeddings mode for semantic search and an optional reranking step to improve result relevance. By default, Inked uses straightforward text matching for speed, and can be extended with embeddings to understand context and meaning across memories. The server stores data locally, and the provided CLI/JS tool writes and reads memories through the Read and Write commands exposed by the MCP interface. To use Inked, add it to your MCP configuration using npx @frgmt/inked, which runs Inked from the npm package.

Once running, you can use the built-in tools to read memories with the read tool (search query or ALL for all memories) and to write memories with the write tool (creating new memories or deleting existing ones). For semantic search, start the server with embedding flags (e.g., --use-embeddings) to load and query against a large language model-backed vector store. For best results, you can combine embedding-based search with the reranking feature (--use-reranking) to re-score top candidates. Note that embedding features are experimental and may have higher memory and latency requirements.

Prerequisites:

• Node.js and npm installed on your system
• Basic familiarity with MCP server configuration

Option A: Global install (recommended for simplicity)

1. Install Inked globally via npm:

npm install -g @frgmt/inked

2. Run Inked as part of your MCP configuration (see mcp_config):

{
  "mcpServers": {
    "inked": {
      "command": "npx",
      "args": ["@frgmt/inked"]
    }
  }
}

Option B: Local development

1. Clone or download Inked source (or use the npm package in your project).
2. Install dependencies and build if required by the project:

git clone https://github.com/frgmt/inked.git
cd inked
npm install
npm run build

3. Run Inked in your MCP config (you can reference the local path if needed). For example:

{
  "mcpServers": {
    "inked": {
      "command": "node",
      "args": ["dist/index.js"]
    }
  }
}

Prerequisites summary: Node.js environment, access to npm, and a MCP-compatible runtime to load the Inked server as shown above.

Tips and considerations:

• The default setup favors fast text search. If you need semantic search, enable embeddings with flags like --use-embeddings (e.g., via the MCP args) and be prepared for higher memory usage.
• Embedding models (e.g., Qwen series) can require several gigabytes of RAM depending on the chosen model and configuration.
• Inked stores data locally under ~/.inked/; ensure appropriate filesystem permissions and backups if you rely on memory history.
• The AI features are experimental; performance and availability may vary, and there may be limitations noted in the official docs or within the server output.
• If you encounter issues with embeddings not loading, verify that the model download completed successfully and that your system meets the required RAM and disk space.
• When using MCP, ensure the Inked service is reachable by the MCP runtime and that the command/args align with your environment (npx usage may download the package at runtime).