MCP and agents
Give your coding agent evidence, progress reporting, and controlled project tools.
Local MCP
Everock supplies a stdio MCP server in apps/mcp. Log in through the CLI first. The server reuses that session when an explicit CI API key is not supplied.
Configure your MCP client's command and arguments for this checkout:
{
"command": "pnpm",
"args": ["--dir", "/absolute/path/to/everock", "--filter", "@everock/mcp", "start"],
"env": { "EVEROCK_API_URL": "http://localhost:4001" }
}
Your agent client determines the outer configuration format and permission prompts. Restart or reconnect its MCP session after changing the server configuration.
Read evidence
| Tool | Purpose |
|---|---|
list_projects | Find accessible projects |
list_project_components | Select the application and repository |
get_incidents | Find current problems |
get_incident_context | Load evidence, comments, affected sessions, replays, changes, and user impact |
get_requests | Search recent request logs |
get_telemetry_usage | Read account counters and protective limits |
get_event_original | Read accepted event data with archive verification |
get_incident_context also records an active analysis phase. It is not a completely passive read: Activity shows that the agent started investigating.
Report work
Use add_incident_note to publish findings and update_incident_progress for analyzing, proposing, or correcting states. Finish each phase as completed or failed. Resolve the incident after verifying the underlying fix.
The tool set also includes proposals and hosting operations. Your client and account permissions govern execution; an MCP connection does not silently authorize every mutation.
Tune project settings
The owner can let an agent read and adjust detection thresholds or the alert rules of existing manual monitors. Use agent-managed settings for the available tools, defaults, and the difference between stopping notifications and stopping HTTP checks.
Remote transport
The repository also contains an HTTP MCP transport with OAuth identity validation. It requires its own public HTTPS URL, authorization-server configuration, and server credentials. Do not assume a local stdio session creates a permanently online cloud agent.
Everock does not supply an internal repair model. Your own agent and its tools perform the investigation and repair.