lcc plugin for ChatGPT)Effective date: 1 October 2026
Applies to: the “Context Compiler” plugin (lcc) hosted at
https://lcc-production.up.railway.app/mcp, and the lcc open-source project it is
built from (https://github.com/lucasmartins-ai/lcc).
The plugin processes the text you paste into a chat on the server that runs it, in memory, for the length of a single request. It is not stored, not logged as content, and not sent to any third party by default. The developer receives no copy of your data.
The text you send, and the goal you state with it. These arrive in the request body of a tool call. They exist in server memory only while the request is being processed.
Nothing else. The plugin does not collect or request:
Sole purpose: score each block of the supplied text against the supplied goal, drop the blocks the scoring finds irrelevant, and return the remainder verbatim plus a report naming what was dropped. There is no other use.
No third party, by default. The hosted service runs the mechanical provider, which is
entirely local to the server process. No model API is called and no text leaves the
server.
One exception exists and is not enabled on the hosted deployment: the optional jev
provider sends your text to TypeSafe (https://api.typesafe.ai/v1/systemone) for
semantic scoring, and the optional nimble provider uses a locally hosted model. Neither
is configured on the hosted instance — it requires an explicit TYPESAFE_API_KEY that the
deployment does not set. If that ever changes, this policy must be updated first.
No retention. Text is held in process memory for the duration of the request and is discarded when the response is returned. The service writes no database, no user files, and no content-addressed store.
The server’s operational log records the request method, path, and status code only
(for example POST /mcp 200). It does not record request bodies, tool arguments, or
response contents.
The open-source CLI and MCP server have an opt-in request log (LCC_REQUEST_LOG=1) that
stores redacted goal strings locally, never the context. It is off by default and not
enabled on the hosted deployment. If that ever changes, this policy must be updated first.
None in the request path. The plugin does not embed trackers, analytics, or advertising SDKs, and it makes no outbound requests while serving a tool call.
The plugin is not directed at children under 13 and collects no personal information from anyone, of any age.
Requests are served over HTTPS only. The plugin’s tools are read-only: they compute and
return text and never modify files, accounts, or external systems. See
SECURITY.md in the
project for the underlying local-first security model.
Depending on where you live, you may have rights to access, correct, delete, or port your personal data, and to object to or restrict processing. Because the plugin retains nothing, these rights are satisfied by default: there is no data about you to access, correct, delete, or export. For any privacy question, open an issue at https://github.com/lucasmartins-ai/lcc/issues.
Material changes to this policy will be published at this URL before taking effect, with the effective date updated above.