Doclea

You run the Doclea MCP Server to provide persistent memory for your AI coding assistants. It stores memories, decisions, patterns, and other project knowledge to improve code understanding, recommendations, and collaboration across your workflow.

How to use

Use the MCP client inside your coding environment to interact with Doclea memory services. You can store memories, search your past decisions, generate changelogs, and map codebase expertise to identify who should review changes. The standard flow is to start the local MCP server, connect your MCP client, and then run memory-related commands as part of your daily development tasks.

How to install

Prerequisites you need before starting: Bun v1.0+ (for running the MCP), Docker and Docker Compose (for services like Qdrant and embeddings), and Git to fetch the code.

# Step 1: Clone the project and navigate to the MCP package
git clone https://github.com/your-org/doclea.git
cd doclea/packages/doclea-mcp

# Install dependencies
bun install

# Download embedding model (first time only, ~130MB)
./scripts/setup-models.sh

# Build the MCP server
bun run build

Step 2: Start the required services and verify they are ready.

# Start Qdrant and Embeddings services
bun run docker:up

# Verify services
curl http://localhost:6333/readyz   # Should return "ok"
curl http://localhost:8080/health   # Should return "ok"

Configuration

Create a project configuration file to tailor embedding, vector database, and storage settings. You can adjust endpoints and paths as needed.

{
  "embedding": {
    "provider": "local",
    "endpoint": "http://localhost:8080"
  },
  "qdrant": {
    "url": "http://localhost:6333",
    "collectionName": "doclea_memories"
  },
  "storage": {
    "dbPath": ".doclea/local.db"
  }
}

Starting the MCP server from a local build

The following start configurations assume you are running the MCP server locally after building it. You can run the server directly via Bun for development or production use.

Troubleshooting

If the MCP server doesn’t appear in your MCP client, verify the absolute path to the server executable in your client configuration, ensure the build finished successfully, and restart the client if needed.

Development

For development and testing, you can run in hot-reload mode and execute unit/integration tests as you iterate on the MCP server.

# Run in development mode with hot reload
bun run dev

# Run tests
bun test

# Run unit tests only
bun run test:unit

# Run integration tests (requires Docker services)
bun run test:integration

# Type check
bun run typecheck

# Build for production
bun run build

Architecture

The Doclea MCP Server integrates with Claude Code to provide memory tools, knowledge indexing, and bootstrap capabilities. It connects to a local vector store (Qdrant), an embeddings service, and a SQLite-based metadata store, enabling fast semantic search and memory management during development.

Available tools

doclea_store

Store a memory entry such as a decision, solution, pattern, architecture, or note.

doclea_search

Perform semantic searches across memories to retrieve relevant context.

doclea_get

Retrieve a memory by its unique identifier.

doclea_update

Update an existing memory entry.

doclea_delete

Delete a memory by its ID.

doclea_commit_message

Generate a conventional commit message from staged changes.

doclea_pr_description

Create a PR description with contextual details.

doclea_changelog

Generate a changelog between refs for developers or users.

doclea_init

Initialize a project by scanning git history, docs, and code to bootstrap memories.

doclea_import

Import memories from Markdown files or ADRs.