RetrySight The usual workflow for checking agent cost is a context switch: stop coding, open a dashboard in a browser tab, find the right session, come back. RetrySight collapses that loop by exposing its data over MCP — so the same tool that produced the retries can also answer questions about them, in the same window you’re already working in.
What this looks like
With RetrySight’s MCP server wired into Cursor, VS Code, or Kiro, you can ask things like:
- “How many retries happened on this file today, and what kind?”
- “Summarize this session’s token spend before I close it out.”
- “Which of my collectors haven’t sent a heartbeat recently?”
The client resolves these against real fleet data — task history, retry classification, and collector status — the same signals covered in the retry taxonomy, returned as structured MCP tool calls rather than a dashboard screenshot.
Lite: local, loopback-only
RetrySight Lite runs an MCP server against its own bundled backend on 127.0.0.1:18081 — nothing leaves your machine. Packaged installs generate ready-made MCP JSON from the app itself (open MCP in Lite and copy it in); a dev clone points at retrysight-lite/manager and an admin token instead. Full steps: Lite MCP setup.
Cloud and Enterprise: fleet-wide
- Cloud mints a long-lived MCP token against your account and talks to the hosted manager over the same origin as the dashboard — useful tools include dashboard summaries, task lookups, and fleet-wide collector status. Setup: Cloud MCP.
- Enterprise Integration+ runs the equivalent MCP server against your own on-prem manager, so fleet-wide queries stay inside your network. See installation tiers.
In every case it’s the same underlying data model — session timelines, retry classes, token spend — just scoped to what that deployment can see: one machine on Lite, one account on Cloud, one fleet on Enterprise.
Try it
- Get RetrySight running — Lite for a single machine, or Cloud/Enterprise for a fleet.
- Wire the MCP config into your editor and restart it.
- Ask a real question about a session you just ran, instead of opening a new tab to find out.