Skip to content
Conduit

mcp server

mcp-local-dev-toolbox

Local MCP server for dev agents. Explicit tool inventory, path allowlists, no generic shell exec.

Set up

Runs on your computer. Pick your agent and operating system.

Before you start

  • Node.js 20 or newer
~/Library/Application Support/Claude/claude_desktop_config.json
{
  "mcpServers": {
    "mcp-local-dev-toolbox": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-local-dev-toolbox"
      ]
    }
  }
}

same format Cursor, GitHub Copilot CLI, Junie, JetBrains AI Assistant, Augment (Auggie CLI), Factory Droid, Kiro, Devin CLI +12

Values in angle brackets are placeholders. Restart Claude Desktop after editing its config.Claude Desktop docs

README

synced Oct 23, 2026

mcp-local-dev-toolbox

A local MCP server that gives AI agents safe, explicit access to filesystem, git, search, and test-runner tools — no generic shell exec.

Setup

git clone https://github.com/yourname/mcp-local-dev-toolbox
cd mcp-local-dev-toolbox
npm install && npm run build

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "local-dev-toolbox": {
      "command": "node",
      "args": ["/absolute/path/to/mcp-local-dev-toolbox/dist/server.js"],
      "env": {
        "WORKSPACE_ROOT": "/absolute/path/to/your/project",
        "ALLOW_WRITES": "false",
        "AUDIT_LOG_PATH": "/tmp/mcp-local-dev-toolbox/audit.jsonl"
      }
    }
  }
}

Set ALLOW_WRITES to "true" when you want the agent to edit files. Omit it for read-only access — git, search, and test-running still work. See examples/cursor-config.json for Cursor setup.

Tools

Full reference: TOOLS.md

ToolCategoryDescription
read_fileFilesystemRead a file's contents
write_fileFilesystemWrite or overwrite a file (requires ALLOW_WRITES)
list_dirFilesystemList directory contents
file_existsFilesystemCheck if a path exists
git_statusGitWorking tree status
git_diffGitDiff against HEAD or a ref
git_logGitCommit history
git_branchGitList or switch branches
run_testsTestsRun test suite, optionally filtered by pattern
get_coverageTestsCoverage summary for the workspace
grep_codebaseSearchRegex search across project files
find_filesSearchGlob-based file finder
search_symbolsSearchSymbol-level search (functions, classes)

Safety model

  • Path allowlist — every filesystem operation is validated against WORKSPACE_ROOT; paths outside it are rejected before the tool runs.
  • No generic exec — there is no run_command tool. Every capability is a named, typed tool with a narrow interface.
  • Audit log — every tool call (allowed or blocked) is appended to AUDIT_LOG_PATH as newline-delimited JSON.

Project layout

src/
  server.ts          # MCP server entry point
  tools/
    filesystem.ts    # read_file, write_file, list_dir, …
    git.ts           # git_status, git_diff, git_log, …
    tests.ts         # run_tests, get_coverage, …
    search.ts        # grep_codebase, find_files, …
  safety.ts          # SafetyChecker — path validation, write guard
  audit.ts           # Append-only audit log
examples/
  claude-config.json
  cursor-config.json
TOOLS.md             # Complete tool reference

Configuration

VariableRequiredDefaultDescription
WORKSPACE_ROOTYes—Absolute path the agent is allowed to access
ALLOW_WRITESNofalseEnable write_file and other mutating tools
AUDIT_LOG_PATHNo~/.mcp-local-dev-toolbox/audit.jsonlWhere to write the audit log

Status

Portfolio/demonstration project. Tool interfaces, safety wiring, and server scaffolding are complete and runnable. Tools marked [stub] in TOOLS.md have full interfaces but delegate to TODO implementations — the real logic would use simple-git, fast-glob, and child_process with argument arrays. PRs welcome.

↑↓ navigate openesc close