diff --git a/CHANGELOG.md b/CHANGELOG.md index 07a5cf7..034901c 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -15,9 +15,9 @@ Product release notes: [docs.proofable.me/changelog](https://docs.proofable.me/c ### Changed -- **The hosted surface carries the trust layer.** `https://mcp.proofable.me/mcp` serves the proof and permission tools (prove something once, reuse the proof, hold an agent to a permission you can revoke), and no longer serves work that runs on the Proofable platform: background and scheduled jobs, the hosted browser, connected apps, a saved provider key, remote MCP servers, web search and page fetch, or the encrypted vault. Those run in Proofable itself, where the run holds its own permission and a human approves side effects; a call to one from a connection answers `platform_only`. Saving or revoking a key happens on your profile, so a credential is never passed through a client. `tools/list` and the published card both stop at the trust layer. +- **Saving a key moves to your profile.** `https://mcp.proofable.me/mcp` no longer accepts a key as a tool argument: `tools/list` does not offer the tool that saved one, and a call to it is rejected as an unknown tool, so a credential is never passed through a client. Every other tool stays on every plan for a signed-in connection, on your credits and within your permissions: the proof and permission tools, jobs, the hosted browser, connected apps, remote MCP servers, session notes, saved tools, web search, page fetch, and your saved keys (list, remove, or call a provider with one, never reading the key back). The published card lists twelve tools and no vault tool. -- **Dedicated ChatGPT plugin endpoint.** The ChatGPT plugin connects to `https://mcp.proofable.me/mcp/openai`, which serves the eight read and guided-verification tools with the same OAuth sign-in; `proofable_verify_or_guide` returns the hosted link whenever a verification step is needed. Cursor, Claude Code, Codex, and every other client see the trust-layer tool list on `/mcp`. +- **Dedicated ChatGPT plugin endpoint.** The ChatGPT plugin connects to `https://mcp.proofable.me/mcp/openai`, which serves the eight read and guided-verification tools with the same OAuth sign-in; `proofable_verify_or_guide` returns the hosted link whenever a verification step is needed. Cursor, Claude Code, Codex, and every other client get the full tool list on `/mcp`. ## [0.1.7] - 2026-10-10 diff --git a/plugins/proofable-mcp/skills/proofable-trust-workflow/SKILL.md b/plugins/proofable-mcp/skills/proofable-trust-workflow/SKILL.md index 4a3b1e5..5c47285 100644 --- a/plugins/proofable-mcp/skills/proofable-trust-workflow/SKILL.md +++ b/plugins/proofable-mcp/skills/proofable-trust-workflow/SKILL.md @@ -46,7 +46,7 @@ Proofable: Blocked. A required trust condition was not satisfied. Next: Do not c - Proofs stay off the chain unless the user asks for a blockchain record. Do not prompt for a wallet otherwise. - Use proof IDs only from tool responses. Prefer “proof ID” / “portable proof” in user text. -- Never ask for a credential and never accept one. A key is saved on the profile in Proofable, where a run uses it without the value reaching a model or an app. This connection carries no tool that stores a secret; a call to one answers `platform_only`. +- Never ask for a credential and never accept one. A key is saved on the profile in Proofable, where a run uses it without the value reaching a model or an app. This connection carries no tool that stores a secret. - Use live verifier IDs from **`proofable_context`** / **`proofable_verifiers_catalog`**. Do not hardcode a second catalog. ## Setup and mount diff --git a/skills/proofable-trust-workflow/SKILL.md b/skills/proofable-trust-workflow/SKILL.md index 4a3b1e5..5c47285 100644 --- a/skills/proofable-trust-workflow/SKILL.md +++ b/skills/proofable-trust-workflow/SKILL.md @@ -46,7 +46,7 @@ Proofable: Blocked. A required trust condition was not satisfied. Next: Do not c - Proofs stay off the chain unless the user asks for a blockchain record. Do not prompt for a wallet otherwise. - Use proof IDs only from tool responses. Prefer “proof ID” / “portable proof” in user text. -- Never ask for a credential and never accept one. A key is saved on the profile in Proofable, where a run uses it without the value reaching a model or an app. This connection carries no tool that stores a secret; a call to one answers `platform_only`. +- Never ask for a credential and never accept one. A key is saved on the profile in Proofable, where a run uses it without the value reaching a model or an app. This connection carries no tool that stores a secret. - Use live verifier IDs from **`proofable_context`** / **`proofable_verifiers_catalog`**. Do not hardcode a second catalog. ## Setup and mount