mcp2cli
Convert MCP servers into standalone Bash-invokable scripts. Use when user wants to make an MCP server usable as bash commands, convert MCP to CLI, or wrap MCP tools for agent use.
$ Installer
git clone https://github.com/majiayu000/claude-skill-registry /tmp/claude-skill-registry && cp -r /tmp/claude-skill-registry/skills/development/mcp2cli ~/.claude/skills/claude-skill-registry// tip: Run this command in your terminal to install the skill
name: mcp2cli description: Convert MCP servers into standalone Bash-invokable scripts. Use when user wants to make an MCP server usable as bash commands, convert MCP to CLI, or wrap MCP tools for agent use.
mcp2cli
Convert any MCP server's tools into standalone Bash-invokable scripts.
When to Use
- User wants to convert an MCP server to CLI tools
- User mentions making MCP tools "agent-ready" or turning an MCP into a "CLI tool"
- User wants to wrap an MCP server for direct invocation
How to Execute
Spawn a subagent using the Task tool with subagent_type: "general-purpose":
Task tool parameters:
- subagent_type: "general-purpose"
- description: "Convert MCP server to CLI tools"
- prompt: <see below>
Subagent Prompt Template
Convert the MCP server "<package-name>" into standalone CLI tools.
## Step 1: Derive Names
Server name derivation:
- Remove @scope/ prefix (e.g., @anthropic-ai/pkg -> pkg)
- Remove @version suffix (e.g., pkg@1.0.0 -> pkg)
- Remove -mcp suffix (e.g., chrome-devtools-mcp -> chrome-devtools)
- Remove mcp- prefix (e.g., mcp-server-fetch -> server-fetch)
## Step 2: Fetch Package Description (in parallel with discovery)
Try to fetch a rich description to use in registration:
**For npm packages:**
```bash
curl -s "https://registry.npmjs.org/<package>" | jq -r '.readme // .description'
Extract the first paragraph after the title (skip badges/headings).
For PyPI packages:
curl -s "https://pypi.org/pypi/<package>/json" | jq -r '.info.description // .info.summary'
Extract the first paragraph after the title.
Use the first meaningful paragraph (>20 chars) or fall back to short description.
Step 3: Discover Tools (Auto-Fallback)
Try npm first:
npx mcporter list --stdio "npx -y <package>@latest" --name <server> --schema --json
If npm fails, try uvx (Python):
npx mcporter list --stdio "uvx <package>" --name <server> --schema --json
Parse the JSON output. Extract all tools with their schemas.
Step 4: Group Tools
Analyze discovered tools and group logically:
- Max 5-6 tools per wrapper
- Dedicated wrappers for complex/high-frequency tools
- Name files: -.js (lowercase, hyphenated)
Step 5: Generate Wrapper Scripts
For each group, generate a Node.js ES module:
#!/usr/bin/env node
import { execSync } from "child_process";
const MCP_CMD = "<mcp-command>"; // The command that worked (npx or uvx)
const SERVER = "<server-name>";
function callMcp(tool, params = {}) {
const paramStr = Object.entries(params)
.map(([k, v]) => `${k}:${JSON.stringify(v)}`)
.join(" ");
const cmd = `npx mcporter call --stdio "${MCP_CMD}" ${SERVER}.${tool} ${paramStr}`;
try {
return execSync(cmd, { encoding: "utf-8", stdio: ["pipe", "pipe", "pipe"] });
} catch (error) {
throw new Error(error.stderr || error.message);
}
}
const args = process.argv.slice(2);
if (args.includes("--help")) {
console.log("Usage: <filename> [options]");
console.log("");
console.log("Options:");
// List ALL parameters from the tool schema with descriptions
process.exit(0);
}
// Parse --key value pairs
const params = {};
for (let i = 0; i < args.length; i++) {
if (args[i].startsWith("--") && i + 1 < args.length) {
const key = args[i].slice(2);
const value = args[++i];
try {
params[key] = JSON.parse(value);
} catch {
params[key] = value;
}
}
}
try {
const result = callMcp("<tool-name>", params);
console.log(result);
} catch (error) {
console.error("Error:", error.message);
process.exit(1);
}
Requirements:
- ES modules (import, not require)
- --help showing ALL parameters with types and descriptions
- Handle complex params (arrays/objects) via --param ''
- Errors to stderr, exit(1)
Step 6: Write Output Files
mkdir -p ~/agent-tools/<server-name>
Create:
- package.json - with "type": "module"
- README.md - document all tools with usage examples (use symlink names without .js extension)
- Wrapper .js scripts - one per tool group
Make executable and create symlinks for direct invocation:
chmod +x ~/agent-tools/<server-name>/*.js
# Use ~/agent-tools/bin if ~/agent-tools/ exists, otherwise ~/.local/bin
if [ -d ~/agent-tools ]; then
BIN_DIR=~/agent-tools/bin
else
BIN_DIR=~/.local/bin
fi
mkdir -p "$BIN_DIR"
for f in ~/agent-tools/<server-name>/*.js; do
ln -sf "$f" "$BIN_DIR/$(basename "$f" .js)"
done
Note: Ensure the bin directory is in the user's PATH.
Step 7: Register to CLAUDE.md
Append to ~/.claude/CLAUDE.md using the fetched description:
### <Server Name>
<fetched description or fallback generic description>
Usage:
```bash
<server>-<tool1> --help
<server>-<tool2> --param "value"
<server>-<tool3> --url "https://example.com"
Full docs: ~/agent-tools/<server-name>/README.md
Note: Replace `<server>-<tool>.js` with actual generated filenames in the registration.
## Report Back
- Package source: npm or PyPI (which succeeded)
- Description: what was fetched
- Tools discovered: count
- Wrappers generated: list of filenames
- Location: ~/agent-tools/<server-name>/
- Registration: confirmed in ~/.claude/CLAUDE.md
Example Invocation
User: "Convert chrome-devtools-mcp to CLI tools"
Spawn Task with the full prompt above, substituting "chrome-devtools-mcp" for .
Repository
