Integrations
MCP server for coding agents
Coding agents can ask WhosePort what is running before they guess. Any MCP client that runs a stdio server works; the tools are read-only.
Setup
| Agent | Command or config |
|---|---|
| Claude Code | claude mcp add whoseport -- whoseport mcp |
| Cursor | Settings → MCP → add whoseport mcp as a stdio server |
| Other MCP clients | Add a stdio server with command whoseport and args mcp |
mcp.json
{
"mcpServers": {
"whoseport": { "command": "whoseport", "args": ["mcp"] }
}
}Tools
Thirteen tools, all read-only. whoseport_diagnose also calls your model provider, so it needs ANTHROPIC_API_KEY, OPENAI_API_KEY or GEMINI_API_KEY in the server's environment.
| Tool | Returns |
|---|---|
whoseport_inspect_environment | The environment graph: services, ports, processes, containers, projects, dependencies, health. Call it first. |
whoseport_get_environment_changes | What changed between snapshots: ports opened or closed, processes started or exited, containers restarted. |
whoseport_list_ports | Listening ports with owning process and PID. |
whoseport_inspect_port | Who owns a port: process, command, cwd, project, framework, container, HTTP health. |
whoseport_list_services | Detected services with kind, ports, runtime and health. |
whoseport_inspect_service | Probe one service over HTTP and report its health. |
whoseport_inspect_process | Command, args, cwd, parent chain, ports and project for a PID. |
whoseport_list_containers | Containers with state, health, image, published ports and compose labels. |
whoseport_inspect_container | Detailed state of one container. |
whoseport_get_recent_logs | Recent, redacted log lines for a service: container logs or its open *.log files. |
whoseport_check_http | GET or HEAD a local URL: status, latency and a short body excerpt. |
whoseport_find_dependencies | Inferred service-to-service links (API → Postgres via DATABASE_URL) with status. |
whoseport_diagnose | A bounded investigation that returns a diagnosis with cited evidence, and in diagnose mode a proposed plan that is never executed. |
What the agent sees
transcript
you › The API returns 502 on /orders. Investigate. agent › whoseport_inspect_port { port: 8080 } agent › whoseport_diagnose { question: "why is /orders 502" } ↳ 5 evidence items · root cause · confidence high agent › DATABASE_URL points at :5433 but PostgreSQL is on :5432 (ev_1041, ev_1042). I'll change the port in apps/api/.env and restart the API.
The MCP server never changes anything
It exposes no mutating tools, and its investigations run at autonomy level 1. Any fix is made by your coding agent, under that agent's own permission prompts, or through an investigation you approve in the dashboard.
The MCP server collects on your machine and returns the same redacted, normalized evidence the dashboard receives. Nothing is uploaded to WhosePort. See Redaction and privacy.