MCP server
Let your own AI client read a run and propose scenarios.
MCP (Model Context Protocol) is a standard plug that gives an AI client tools. luora-mcp.mjs gives Claude Code, Claude Desktop or Cursor Luora's view of your run: decoded traces, the contract map, findings. Without it, the AI reads raw forge output or a large JSON; with it, it asks Luora for exactly what it needs.
Set it up
curl -fsSLO "https://www.luorasec.xyz/exporter/luora-mcp.mjs" claude mcp add luora -- node /abs/path/luora-mcp.mjs --data=/abs/path/luora-data.json
Tools
| Tool | Access | Returns |
|---|---|---|
| overview | read | Contract, tests, finding counts, coverage and what scenarios may use |
| get_trace(test) | read | One test's decoded frames |
| get_atlas | read | Every mapped piece and what each entry point reads, writes, calls and emits |
| get_piece(id) | read | One piece: source lines, links, findings, reverts, coverage |
| list_findings | read | Slither, Aderyn and Luora findings, by severity or tool |
| propose_scenario | append | Checks a scenario and saves it to luora-scenarios.json |
Proposals in the Lab
A proposal shows up in Atlas → Scenarios under *Proposed by your AI client*: in a local Luora, or on luorasec.xyz while node luora-bridge.mjs runs in the project (switch back to the tab to pick up a new one). Pick it, check it, Run, then export the .t.sol.
The server runs on your machine over stdio, makes no network calls, reads luora-data.json and only writes luora-scenarios.json. Scenarios test intended behaviour only.