MCP server for efficient codebase exploration by coding agents. List exports, get file outlines, read targeted regions, and find references — without loading whole files into context.
| Tool | Input | Output | When to use |
|---|---|---|---|
list_exports |
path (file/folder), filter?, depth? | Name + signature + line + totalLines | Discover what a file/folder contains |
file_outline |
path (file), exportedOnly? | All symbols with line ranges | Understand structure before reading |
read_region |
path, start, end (max 200 lines) | Numbered source lines | Read only the code you need |
find_references |
symbol, path (dir), filePattern?, maxResults? | File:line + context | Trace call chains and usage |
1. list_exports(path, filter) → discover what exists and where
2. file_outline(file) → understand structure and line ranges
3. read_region(file, start, end) → read only the relevant ±30 lines
4. find_references(symbol, dir) → trace who calls it before modifying
No local install needed. Point your MCP client at:
command: npx
args: ["-y", "@dothinh/code-intel-mcp"]
Add to ~/.codex/config.toml:
[mcp_servers.code-intel]
command = "npx"
args = ["-y", "@dothinh/code-intel-mcp"]Add to .cursor/mcp.json (project-level) or global MCP settings:
{
"mcpServers": {
"code-intel": {
"command": "npx",
"args": ["-y", "@dothinh/code-intel-mcp"]
}
}
}claude mcp add code-intel -- npx -y @dothinh/code-intel-mcpUse the same npx command and args in your client's MCP server configuration. The server communicates over stdio using newline-delimited JSON-RPC.
- Node.js >= 20
rg(ripgrep) on PATH — used byfind_references
TypeScript, JavaScript, Vue SFC, and Svelte for export detection and file outlines. find_references works with any text file that ripgrep can read (Python, Go, Rust, etc.).
git clone git@github.com:dothinh115/code-intel-mcp.git
cd code-intel-mcp
yarn install
yarn build
node dist/index.js # starts MCP server on stdioMIT