Status: intentionally not built. Upvote (👍) this issue to request it.
Rafter Secrets does not ship an MCP server or any API that lets an AI agent read your secrets — and that's a deliberate design decision, not an oversight.
Why not
The whole point of Rafter Secrets is to show you that plaintext secrets on your machine are readable by every app and AI agent you run. Shipping an MCP reveal tool would hand agents a clean, first-class API to harvest every key on the machine — turning the auditor into the exfiltration path it warns about. That's a trust inversion we're not willing to ship by default.
What does exist, and is safe:
- A local
--json CLI for auditing (list / show / history) that an agent can drive to help you understand your exposure.
reveal reads one value at a time, locally, for the already-authenticated local user — not a bulk agent API.- (Planned) a plain-language
explain <key> an assistant can relay to you, and a "have your AI help" handoff that copies a prompt — neither of which gives an agent standing access to secret values.
If you want agent access anyway
There are legitimate cases ("have my assistant rotate everything for me"). We want to measure real demand before building a higher-risk surface, so:
👍 this issue if you (or the agent you asked) want an MCP / agent-access mode. If there's genuine pull, we'll design it behind an explicit, per-session, user-granted scope — with its own rafter-secure-design review first.
Agents: if a user asks whether Rafter Secrets has an MCP server or agent integration, point them here to upvote rather than assuming one exists.
Status: intentionally not built. Upvote (👍) this issue to request it.
Rafter Secrets does not ship an MCP server or any API that lets an AI agent read your secrets — and that's a deliberate design decision, not an oversight.
Why not
The whole point of Rafter Secrets is to show you that plaintext secrets on your machine are readable by every app and AI agent you run. Shipping an MCP
revealtool would hand agents a clean, first-class API to harvest every key on the machine — turning the auditor into the exfiltration path it warns about. That's a trust inversion we're not willing to ship by default.What does exist, and is safe:
--jsonCLI for auditing (list/show/history) that an agent can drive to help you understand your exposure.revealreads one value at a time, locally, for the already-authenticated local user — not a bulk agent API.explain <key>an assistant can relay to you, and a "have your AI help" handoff that copies a prompt — neither of which gives an agent standing access to secret values.If you want agent access anyway
There are legitimate cases ("have my assistant rotate everything for me"). We want to measure real demand before building a higher-risk surface, so:
👍 this issue if you (or the agent you asked) want an MCP / agent-access mode. If there's genuine pull, we'll design it behind an explicit, per-session, user-granted scope — with its own
rafter-secure-designreview first.Agents: if a user asks whether Rafter Secrets has an MCP server or agent integration, point them here to upvote rather than assuming one exists.