Introduction
sremcp is a read-only AI SRE for your whole team. It investigates with the MCP servers you connect, calls only the tools you approve, and gives each role exactly the access it needs.
sremcp is an investigation agent for incidents, latency, deploys and capacity. It works through MCP servers: GitHub, Grafana, Kubernetes, Amazon EKS, PostgreSQL, Prometheus, Sentry, or your own. You connect a server once, approve the tools the agent may call, and choose which roles may use it. Then SREs, managers, developers and QA ask questions in plain language, and every tool call is recorded.
Quickstart
Create your organization, connect a first MCP server and run your first investigation, in about ten minutes.
Connect an MCP server
The five steps every server goes through: register, connect, test, approve, grant and activate.
Integrations
Step-by-step guides for GitHub, Grafana, Kubernetes, Amazon EKS, PostgreSQL, Prometheus and Sentry.
Roles & permissions
What Owners, Admins, SREs, Managers, Developers, QA and Viewers can each see and do.
How sremcp works
- An SRE or admin onboards an MCP server: its URL, how it signs in, and its credentials, which are encrypted and never shown again.
- sremcp discovers the server's tools. Every tool starts pending. The agent can't call a tool until an SRE or admin approves it.
- The server is granted to roles. Owners, Admins and SREs use every server. Managers, Developers, QA and Viewers use only the servers granted to their role, for example QA only staging.
- Your team investigates. The agent picks from the approved tools of the servers the person's role may use, gathers evidence, and answers with a summary, the evidence, the likely cause and next steps.
- Everything is recorded. Every answer lands in Runs, and every tool call and change in the Audit log.
Read-only by design
sremcp is built to look, not touch. The agent is instructed never to change anything, and it can only call the tools an SRE or admin approved. That approval is the real boundary, so approve read tools only, and give each server a read-only credential. If you approve a tool that can change something, the agent is able to call it.
Every integration guide shows how to run its server in read-only mode and with least-privilege credentials.
Need help?
Start with the FAQ and Troubleshooting. If you're still stuck, contact us from sremcp.com.