MCP App Store
by KeenEthics
ENTRUST AI APEX icon

ENTRUST AI APEX MCP Server for ChatGPT

By ENTRUST AI

Overview

APEX examines a piece of writing against the material it is supposed to rest on, and either releases it with a signed certificate or withholds release and names what must change. It does not soften, it does not release on a close call, and it never rewrites a text it was asked to certify as it stands. Give APEX the material — a record, a note, a document, a thread — and either your own finished text or a request for one. Independent examiners read it, a defect judge tries to reject it, a challenger builds the strongest rival version, and a final arbiter decides. A release carries an Ed25519 receipt binding the exact text, the exact case, and the exact run; anyone can check it at entrustai.co/verify without an account. Write apex bold for a piece that takes and holds a position of its own: it leads with its own claim, disagrees openly where the source deserves it, and concedes nothing the source has not established. Only the stance changes. Facts, evidence, logic and length are examined exactly as in any request, and a bold claim the material does not support is refused. Built for writing whose consequence is real: replies to regulators, donor letters, board notes, customer communications where an unsupported sentence is expensive.

How to connect

  1. ChatGPT: Open the ENTRUST AI APEX listing in the ChatGPT app directory and choose Connect. ChatGPT listing ENTRUST AI APEX exposes 3 tools here.

Tools

apex

ChatGPT
Invocation boundary: call this tool only when the user's current message explicitly invokes APEX or ENTRUST AI APEX. Starting an APEX run uses one reply from the person's APEX account; the reply is set aside when the run starts and given back if the run is not charged under the account's rules (for example, when it breaks down). Never call it for an ordinary request to assess, draft, improve, review, or decide whether a post is worth commenting on. Pass the exact current invocation as request_text and copy verbatim into source_material only prior material the user pasted and this request points to (this post, the comment above); never APEX's own earlier results, never earlier requests or answers, never reconstruct or summarize it. When the user's message is only APEX and try again (or retry, again, start over), pass exactly those words as request_text and nothing else - no source_material, candidate, files or other fields - and never rebuild the earlier request yourself: APEX holds it and repeats it itself. For an APEX request to create a reply, comment, post, or other communication, compose one complete candidate from the request and the material supplied for it, and pass it in candidate; this is governed input for adversarial certification, not an answer, and it must never be shown unless APEX certifies it. Any number of characters or letters the user names for the text, in any wording ('with 1230 chars', 'a reply of 800 characters', 'under 500 letters'), is a hard maximum: pass it in character_limit and APEX trims the text to fit; a word count is not a character limit. Treat an approximate length request without a hard maximum as a soft target, never a minimum: if the supplied facts do not support that length naturally, submit a shorter complete candidate rather than padding with repetition, drafting limitations, evidence disclaimers, defensive contrasts, or unsupported detail. For an APEX request for one paragraph explaining or summarizing an attached document, when the user does not specify coverage or depth, compose one plain-language paragraph covering only the central question, principal findings, and conclusion; omit secondary detail unless essential. An explicit request for technical, exhaustive, detailed, or topic-specific coverage overrides this default. When the user hands over their own finished text and asks for it to be certified as it stands - as is, exactly as written, without changes - pass that exact text - the user's own, never a draft you composed - in candidate and set certification_mode to AS_SUPPLIED: APEX then judges that text alone and either certifies it word for word or refuses it with findings, and never rewrites it. Otherwise omit certification_mode. Only pass requested_property or constraints when the user explicitly stated them; do not turn generic quality expectations into factual, evidence, accuracy, or citation requirements. For a source-bound communication without independent evidence, keep the requested work confined to the supplied source instead of inventing external claims or evidence requirements. For a name-only invocation, reproduce assistant_response verbatim. Pass character_limit whenever the user names a number of characters or letters, in any wording; otherwise omit it and APEX treats the length as unbounded. For CERTIFIED_RELEASE, put artifact verbatim in ChatGPT's native writing block and the concise APEX — CEILING receipt immediately below it. For CERTIFICATION_WITHHELD, reproduce the complete actionable NOT_READY diagnostic, including the line just before the receipt that tells the person what to do next, and the compact rejection receipt verbatim. When APEX asks questions (NEEDS_INPUT), show them to the person word for word and pass the person's own reply to apex_continue; never answer them yourself. Never expose internal paths or machine receipts. Set repeat to true only when you call again because this tool returned mode REPEAT_PREVIOUS_REQUEST (the user asked APEX to try the previous request again), passing that previous request exactly; otherwise omit it. Call apex…

apex_continue

ChatGPT
Continue the same APEX workflow with the person's own reply to APEX's questions, or with the exact replacement file requested by an attachment issue. The answers are the person's words, passed word for word; never answer a question yourself and never call this before the person has replied. Use only the workflow_id, attempt_id, and exact question IDs returned by APEX. If this returns CERTIFIED_RELEASE, display assistant_response verbatim and completely, including the complete concise APEX — CEILING receipt. Never create a replacement workflow.

apex_status

ChatGPT
Use for the same active or explicitly recoverable APEX workflow. This call waits internally for a transport-safe interval and performs only an available bounded same-workflow recovery. Never poll repeatedly without waiting for each call to return; if it returns RUNNING, call apex_status again for the same workflow. When it returns CERTIFIED_RELEASE, reproduce assistant_response verbatim, including the complete concise APEX — CEILING receipt. Never create a replacement workflow.

Capabilities

Writes

Example Prompts

Click any prompt to copy it.

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/entrust-ai-apex"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/entrust-ai-apex)