MCP server

Run csp as an MCP server so agents can search any local or remote codebase directly.

csp can run as an MCP server so agents can search any codebase directly. Repos are cloned and indexed on demand. The server keeps a hot in-memory cache for the session and shares the same on-disk cache at ~/.csp/index/ as the CLI, so an index built once is reused across both. Local paths are watched for file changes and re-indexed automatically; on-disk reuse is invalidated by source content hash.

The server is launched with bunx @pleaseai/csp mcp. Note the mcp subcommand: semble uses the bare binary, csp does not.

Setup

claude mcp add csp -s user -- bunx @pleaseai/csp mcp

Or install the plugin, which also adds the csp-search sub-agent.

Add to ~/.cursor/mcp.json (or .cursor/mcp.json in your project):

~/.cursor/mcp.jsonjson

{
  "mcpServers": {
    "csp": {
      "command": "bunx",
      "args": ["@pleaseai/csp", "mcp"]
    }
  }
}

Add to ~/.codex/config.toml:

~/.codex/config.tomltoml

[mcp_servers.csp]
command = "bunx"
args = [
  "@pleaseai/csp",
  "mcp"
]

Add to ~/.opencode/config.json:

~/.opencode/config.jsonjson

{
  "mcp": {
    "csp": {
      "type": "local",
      "command": ["bunx", "@pleaseai/csp", "mcp"]
    }
  }
}

Add to .vscode/mcp.json in your project (or your user profile’s mcp.json):

.vscode/mcp.jsonjson

{
  "servers": {
    "csp": {
      "command": "bunx",
      "args": ["@pleaseai/csp", "mcp"]
    }
  }
}

Add to ~/.copilot/mcp-config.json:

~/.copilot/mcp-config.jsonjson

{
  "mcpServers": {
    "csp": {
      "command": "bunx",
      "args": ["@pleaseai/csp", "mcp"]
    }
  }
}

Add to ~/.codeium/windsurf/mcp_config.json:

~/.codeium/windsurf/mcp_config.jsonjson

{
  "mcpServers": {
    "csp": {
      "command": "bunx",
      "args": ["@pleaseai/csp", "mcp"]
    }
  }
}

Add to ~/.gemini/settings.json:

~/.gemini/settings.jsonjson

{
  "mcpServers": {
    "csp": {
      "command": "bunx",
      "args": ["@pleaseai/csp", "mcp"]
    }
  }
}

Add to ~/.config/zed/settings.json (or .zed/settings.json in your project):

~/.config/zed/settings.jsonjson

{
  "context_servers": {
    "csp": {
      "command": "bunx",
      "args": ["@pleaseai/csp", "mcp"]
    }
  }
}

Tools

Tool	Description
`search`	Search a codebase with a natural-language or code query. Pass `repo` as a local directory path or an https:// git URL.
`find_related`	Given a file path and line number, return chunks semantically similar to the code at that location.

Both tools accept max_snippet_lines to cap the code returned per result. It defaults to 10, a signature-plus-first-lines preview that lets an agent confirm a location cheaply, then navigate to the file for full context. Pass 0 for the location only, or null for the full chunk when the preview lacks context.

Both tools also accept content (code, docs, config, or all) to choose what a single call searches, for example content: "docs" to look up a guide in a repo the server otherwise indexes as code. It defaults to the server’s configured content (below). Each distinct content selection is indexed and cached separately for the session.

Content selection

By default the MCP server indexes only code files. To also index documentation, config, or everything, append --content docs, --content config, or --content all to the server command, or a combination such as --content code docs. For example, in Claude Code:

claude mcp add csp -s user -- bunx @pleaseai/csp mcp --content all