An experimental Python-based server that locally indexes codebases using ChromaDB and provides a semantic search tool via an MCP (Model Context Protocol) server for tools like Cursor.
-
Clone and enter the repository:
git clone <repository-url> cd cursor-local-indexing
-
Create a
.env
file by copying.env.example
:cp .env.example .env
-
Configure your
.env
file:PROJECTS_ROOT=~/your/projects/root # Path to your projects directory FOLDERS_TO_INDEX=project1,project2 # Comma-separated list of folders to index
Example:
PROJECTS_ROOT=~/projects FOLDERS_TO_INDEX=project1,project2
-
Start the indexing server:
docker-compose up -d
-
Configure Cursor to use the local search server: Create or edit
~/.cursor/mcp.json
:{ "mcpServers": { "workspace-code-search": { "url": "http://localhost:8978/sse" } } }
-
Restart Cursor IDE to apply the changes.
The server will start indexing your specified projects, and you'll be able to use semantic code search within Cursor when those projects are active.
- Open a project that you configured as indexed.
Create a .cursorrules
file and add the following:
<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>
- Start using the Cursor Agent mode and see it doing local vector searches!