Plugging an external tool into a coding agent looks simple until the server refuses to connect and leaves you with no clue why. In Verboo Code the path is short: one command to add it, an approval rule that depends on scope, and a dedicated diagnostic command for when something breaks.
How do you add an MCP server to Verboo Code?
With verboo mcp add <name> <command-or-url>. For a stdio server (a local process):
verboo mcp add sentry -- npx @sentry/mcp-server
For a remote HTTP server, declare the transport:
verboo mcp add --transport http sentry https://mcp.sentry.dev/mcp --header "Authorization: Bearer YOUR_TOKEN"
Four flags control where the config is saved and how the server authenticates:
| flag | what it does |
|---|---|
-s, --scope | local, project or user. Default is local: only the current project, on your machine, sees the server |
-t, --transport | stdio, sse or http. Without this flag Verboo Code assumes stdio |
-e, --env | environment variable for the process, e.g. -e API_KEY=xxx |
-H, --header | HTTP header for a remote server, e.g. -H "Authorization: Bearer ..." |
Does every server you add require approval?
No. Only project scope, because that config gets committed to the repository, so anyone on the team who clones the project inherits the servers in it. That is why a project-scoped server starts out with pending status until someone explicitly approves it. local scope, the default for mcp add, skips that screen entirely: it is your own config, on your own machine, under your own responsibility.
In practice: if you are testing an MCP server on your own, leave it at local scope. Only push it to project once the whole team needs the same server, because that is what triggers the approval flow.
The server will not connect. How do you diagnose it?
With verboo mcp doctor <name>. It reads the config, checks precedence across scopes, and by default attempts the connection for real:
verboo mcp doctor sentry
Careful: without --config-only, doctor may spawn a local process and contact a remote server for real. Only run it in a directory you trust, exactly as the command's own description warns.
| flag | what it does |
|---|---|
--config-only | spawns no process and contacts no remote server, only analyses the configuration |
--json | returns the report as JSON, for scripts or CI |
-s, --scope | restricts the analysis to one scope: local, project, user or enterprise |
Running it with no argument (verboo mcp doctor) diagnoses every configured server at once, useful when you are not sure which of several plugged-in MCPs is the broken one.
How do you enable or disable a server without removing it?
Inside a session, with /mcp enable <name> or /mcp disable <name>. Passing all instead of a name toggles every server at once, which helps isolate whether a problem comes from one specific MCP or from something else.
A tool plugged in correctly saves the context that would otherwise go into troubleshooting the plumbing instead of the actual problem. That is the kind of detail that separates a smooth session from a stuck one. Meet Verboo Code.



