Workspace Code Index

You run a local code indexing MCP server that builds a semantic, code-aware search index from your projects and exposes it to MCP clients like Cursor. This server indexes code with ChromaDB locally and lets you query your codebase contextually, enabling fast, relevant results without leaving your editor.

How to use

To use this MCP server with a client like Cursor, point the client to the local MCP endpoint and let it query the indexed codebase. You will enable semantic search over your projects and receive results that reflect code intent, not just keyword matches.

How to install

Prerequisites you need before starting:

# Ensure you have Docker and docker-compose installed
# 1) Clone the project
# 2) Set up environment variables
# 3) Start the indexing service

Step-by-step commands you will run:

git clone <repository-url>
cd cursor-local-indexing

cp .env.example .env

Edit the environment to point to your codebase and folders to index.

# Example values to place in .env
PROJECTS_ROOT=~/projects
FOLDERS_TO_INDEX=project1,project2

Start the indexing service using Docker Compose.

docker-compose up -d

Configure your MCP client (Cursor) to connect to the local MCP server by creating or updating the MCP configuration used by Cursor.

{
  "mcpServers": {
    "workspace-code-search": {
      "url": "http://localhost:8978/sse"
    }
  }
}

Restart Cursor to apply the new MCP target.

Additional sections

Project indexing and usage notes you should know:

• The server will index the folders you specify in .env and keep the index locally for fast semantic search.

• You can open an indexed project in Cursor and the local search will guide queries with vector-based results.

• You can customize how Cursor asks the assistant to search by adding a cursor rules file that instructs the agent to use the code search tool first.

Example rule that instructs the agent to prefer code search before other commands:

<instructions>
For any request, use the @search_code tool to check what the code does.
Prefer that first before resorting to command line grepping etc.
</instructions>

Available tools

search_code

Code search tool used to inspect what the code does and retrieve context from the indexed code.

workspace_code_search

MCP endpoint that provides semantic search over locally indexed code via HTTP.

cursor_agent

Cursor Agent mode performs local vector searches against the indexed data to surface relevant code snippets.