Any Script MCP Server

How to use

Register your toolset with an MCP client configuration, then run MCP Tools from any client that supports MCP. Each tool is defined with input parameters and a run script. Inputs become environment variables or JSON payloads, and the tool’s run script executes in a shell with those inputs available for processing.

How to install

Prerequisites: you need Node.js and npm to use the MCP server via npm or npx.

Install and run the MCP server using npx. You can also register the server through a management CLI if you are using a cluster or registry workflow.

$ claude mcp add any-script \
  -s user \
  -- npx any-script-mcp
```

```json
{
  "mcpServers": {
     "any-script": {
       "command": "npx",
       "args": ["any-script-mcp"]
     }
  }
}

Additional setup notes

Create a configuration file at the standard config path or specify a custom path using an environment variable. This server supports loading multiple configuration files and merging their tools.

Configuration file paths and environment variable examples shown here illustrate how to point to a single file or multiple files, so you can separate common tools from personal or project-specific customizations.

You can test your configuration with the MCP Inspector tool, which opens a web interface to view registered tools and test them interactively.

Example configuration

# yaml-language-server: $schema=https://raw.githubusercontent.com/izumin5210/any-script-mcp/main/config.schema.json
tools:
  - name: echo
    description: Echo a message
    inputs:
      message:
        type: string
        description: Message to echo
    run: |
      echo "Received: $INPUTS__MESSAGE"
      
  - name: git_status
    description: Check git status with optional branch
    inputs:
      branch-name:
        type: string
        description: Branch to check out
        required: false
      verbose:
        type: boolean
        description: Show verbose output
        default: false
    run: |
      if [ -n "${INPUTS__BRANCH_NAME:-}" ]; then
        git checkout "$INPUTS__BRANCH_NAME"
      fi
      
      if [ "$INPUTS__VERBOSE" = "true" ]; then
        git status -v
      else
        git status
      fi
  
  # Delegate search to codex CLI. Inspired by https://github.com/yoshiko-pg/o3-search-mcp
  - name: codex-search
    description: AI agent with web search for researching latest information, troubleshooting program errors, discussing complex problems and design decisions, exploring advanced library usage, and investigating upgrade paths. Supports natural language queries.
    inputs:
      prompt:
        type: string
        description: What you want to search, analyze, or discuss with the AI agent
    run: |
      codex exec \
        --model gpt-5 \
        --sandbox workspace-write \
        --config "sandbox_workspace_write.network_access=true" \
        "$INPUTS__PROMPT" \
        --json \
        | jq -sr 'map(select(.msg.type == "agent_message") | .msg.message) | last'
    timeout: 600000  # 10 minutes for complex AI operations
      
  - name: build
    description: Run build process with tests
    run: |
      npm run build
      npm test
    timeout: 180000  # 3 minutes for build and test

Security and best practices

Limit the tools exposed by each configuration to those you explicitly define. Use input validation within each tool’s run script and prefer using environment variables or INPUTS_JSON to pass structured data to scripts. Keep sensitive data out of logs by avoiding echoing raw secrets and consider restricting tool execution to trusted environments.

echo

Echo a message to the console using a defined input

git_status

Check git status with optional branch and verbosity controls

codex-search

AI agent that performs web search and returns synthesized results for a given prompt

build

Run build process and tests to verify project integrity