Unable to set up usage of npm run <command> as a tool

[cbarrerah]

Hi,

First of all, the project is amazing and super useful, so keep up the good work ^.^

I managed to get the claude desktop app talking with the codemcp server and being able to manage files without resorting to the filesystem mcp server and doing an amazing job at helping to manage the coverage improvements in an easy and useful way. But I was thinking that if the codemcp server was able to actually run the tests being implemented to check the execution results, or running the coverage command to actually see the improvements, it would enable for a more autonomous agent behavior.

I have setup some npm run scripts in my package.json file (typescript project using npm) so I am able to execute in the terminal in the root of the project both "npm run test" and "npm run coverage" and it gets the tests running, or the coverage reported in the cli. So I was expecting that by making those commands available as tools to the codemcp server, it would work. I defined the tools as:

[commands]
npm =["/Users/username/.nvm/versions/node/v18.16.0/bin/npm"]
prettier = ["npm", "run", "lint"]
lint = ["npm","run", "prettier" ]
coverage = ["npm", "run", "coverage"]
test = ["npm", "run", "test"]

But when asked about coverage it tries to use the coverage command, but errors out

Error: Error during npm: Command failed with exit code 1: npm run coverage
Stdout:

appName@1.6.31 coverage
vitest run --config ./vitest.config.ts --coverage

Stderr:
⎯⎯⎯⎯⎯⎯⎯ Startup Error ⎯⎯⎯⎯⎯⎯⎯⎯
TypeError: crypto$2.getRandomValues is not a function
at resolveConfig (file:///Users//dev/webdev/appName/node_modules/vite/dist/node/chunks/dep-ByPKlqZ5.js:54232:16)
at async _createServer (file:///Users//dev/webdev/appName/node_modules/vite/dist/node/chunks/dep-ByPKlqZ5.js:43841:18)
at async createViteServer (file:///Users//dev/webdev/appName/node_modules/vitest/dist/chunks/cli-api.BZbq1zTX.js:10310:18)
at async createVitest (file:///Users//dev/webdev/appName/node_modules/vitest/dist/chunks/cli-api.BZbq1zTX.js:13530:18)
at async prepareVitest (file:///Users//dev/webdev/appName/node_modules/vitest/dist/chunks/cli-api.BZbq1zTX.js:13949:15)
at async startVitest (file:///Users//dev/webdev/appName/node_modules/vitest/dist/chunks/cli-api.BZbq1zTX.js:13880:15)
at async start (file:///Users//dev/webdev/appName/node_modules/vitest/dist/chunks/cac.87F_onld.js:1571:17)
at async CAC.run (file:///Users//dev/webdev/appName/node_modules/vitest/dist/chunks/cac.87F_onld.js:1542:3)

View result from codemcp from codemcp (local)

What am I doing wrong to configure the right command to be executed and available? Is there a different environment being set up when codemcp is running the command, and if so how could I configure it properly?

Thanks :)

[ezyang]

I have had this problem and it usually boils down to npm being in something that is not in the global system PATH at the time Claude Desktop is launched (e.g., the installer script added it to bash_profile, but Claude Desktop never executes that!) An easy way to test if this hypothesis is true is patching the toml to point to the full path of npm, or arrange for it to be installed globally e.g., in /usr/local/bin

For the sanity of users, it might be better if we naughtily grovel in bash_profile to get all envvars we need though...