Connect an AI assistant
Let Claude Code or another MCP client read your Flowrensic findings, episodes and runs over MCP.
Flowrensic has an MCP server, so an AI assistant can answer why a workflow failed or got stuck from what Flowrensic found, and point you to the finding. MCP is the Model Context Protocol, the way AI assistants such as Claude Code call tools.
The MCP server is part of the Flowrensic server. It is at /mcp of the same address as the UI and the API, and speaks streamable HTTP. With the quickstart that is:
http://127.0.0.1:8080/mcpOn AWS it is /mcp of the UI name you gave the Terraform module, such as https://flowrensic.example.com/mcp.
Who can reach it
The same people who can open the Flowrensic UI, and no one else. The UI has no login: who can open it is decided by who can reach its address, such as the IP ranges and security groups you give the Terraform module. The MCP server follows the same rule. It is not on the ingest port that CI jobs send to, and an env access key does not open it. A call a browser sends from a page of another site is refused.
Every tool only reads. No tool changes a finding, a run or a setting.
Claude Code
Add Flowrensic once, with the address of your Flowrensic:
claude mcp add --transport http flowrensic http://127.0.0.1:8080/mcpThen ask, for example:
Why did the payout workflow fail in prod today? Use Flowrensic.Claude Code lists the findings, reads the one that fits and its runs, and gives you the link to the finding in the Flowrensic UI.
Other MCP clients
Most MCP clients take the server as a JSON entry with its address. For a project of Claude Code it goes in .mcp.json at the root of the project:
{
"mcpServers": {
"flowrensic": {
"type": "http",
"url": "http://127.0.0.1:8080/mcp"
}
}
}For another client, see its docs for where the entry goes. The client must speak streamable HTTP.
The tools
| Tool | What it gives |
|---|---|
| list_findings | The findings, the newest first, 20 a page, with the filters of the Findings page: severity, class, env, worker, workflow type, status and archived. Archived findings are left out unless asked for. |
| get_finding | One finding: what it found, its details, the recommended action, its episodes, the observations of one episode a page at a time, such as each failed run, and the remediations people did from the UI. |
| get_run | One run, as Flowrensic stored it: its workflow, env, status and times, a summary of its history, and the findings that hold it. |
Every answer has the link to the same thing in the Flowrensic UI, so you can check what the assistant says. Long text is cut at 300 characters, and the answer says so.