Configuration

The local edition is configured entirely through environment variables — there's no config file to manage. Set them in the environment your MCP client launches the server in (or inline before the command).

Data locations

Variable	Default	Controls
`CLEAN_REPOS_DIR`	`~/.clean/repos`	Where GitHub repos are cloned for indexing.
`CLEAN_DB_PATH`	`~/.clean/metadata.db`	The SQLite metadata store (which repos are indexed, their status).
`CLEAN_PERSIST_PATH`	`~/.clean/index`	The LanceDB vector store.

Example — store the index on a different volume:

CLEAN_PERSIST_PATH=/data/clean/index python -m clean.local.mcp_server

Embedding & indexing

Variable	Default	Controls
`CLEAN_EMBEDDING_MODEL`	`all-MiniLM-L6-v2`	The sentence-transformer model used to embed code (384-dim). A smaller model means faster, cheaper indexing.
`CLEAN_SHOW_PROGRESS_BAR`	`false`	Whether to show an embedding progress bar.

Tooling

Variable	Default	Controls
`CLEAN_DISABLE_INDEX_TOOL`	unset	When set, hides the `index_repo` and `delete_repo` tools (used for benchmarks or read-only deployments).

Setting these in your agent's MCP config keeps them scoped to Clean. For example, in .mcp.json you can add an "env" block alongside "command" and "args".

Worked example

A .mcp.json that pins a default repo and relocates the index:

{
  "mcpServers": {
    "clean": {
      "command": "python",
      "args": ["-m", "clean.local.mcp_server", "--repo", "facebook/react"],
      "env": {
        "CLEAN_PERSIST_PATH": "/data/clean/index"
      }
    }
  }
}

Configuration

Data locations

Embedding & indexing

Tooling

Worked example

On this page