Document a reliable MCP Inspector CLI workflow

[autocarl]

Summary

MCP Inspector CLI is useful for verifying Repl-generated MCP tools/resources/prompts, but direct command-line invocation is easy to get wrong when passing both inspector options and target process arguments.

Observed friction

This kind of invocation can accidentally let --tool-arg consume target command tokens such as dotnet:

npx --yes @modelcontextprotocol/inspector --cli \
  --method tools/call \
  --tool-name ops_deploy_plan \
  --tool-arg service=checkout \
  --transport stdio \
  dotnet run -- mcp serve

A reliable pattern was to add a wrapper script:

#!/usr/bin/env bash
set -euo pipefail
cd "$(dirname "$0")"
exec dotnet run -- mcp serve

Then call:

npx --yes @modelcontextprotocol/inspector --cli ./run-mcp.sh \
  --method tools/list

npx --yes @modelcontextprotocol/inspector --cli ./run-mcp.sh \
  --method tools/call \
  --tool-name ops_deploy_plan \
  --tool-arg service=checkout

Documentation suggestion

Add a short “Verify with MCP Inspector” section that covers both:

1. source/dev mode, using dotnet run or a wrapper script;
2. packaged/compiled mode, using the installed app executable.

Why this matters

This gives coding agents and users a deterministic way to verify the generated MCP surface before configuring an actual client.

[carldebilly]

Other considerations:

• Developers are often on Windows (and/or using Powershell), so the solution should include a Windows story