MCP App Store
by KeenEthics
Energent.ai icon

Energent.ai MCP Server for ChatGPT

By Cambio Corp. (Energent.ai)

Overview

AI assistants can make mistakes in numbers, formulas and claims. Energent gives ChatGPT a second opinion. Ask ChatGPT to send its work to Energent, and an independent Energent auditor checks it against your source files in its own sandbox. The auditor writes a report that shows which numbers and claims are correct, which are wrong, and what to fix. You read the findings in the same chat, and you can open the full audit in the Energent web app. Use it to check financial models, reports, reconciliations, summaries and data analyses before you share them. After you fix the errors, ask for a re-audit: the auditor keeps the earlier findings and checks what changed. Requires an Energent.ai account.

How to connect

  1. ChatGPT: Open the Energent.ai listing in the ChatGPT app directory and choose Connect. ChatGPT listing Energent.ai exposes 9 tools here.

Tools

energent_get_file

ChatGPT
Get one file listed by energent_list_files. A small (<= 256 KB) text, Markdown, CSV, JSON or HTML file comes back inline as text; any other file comes back as a download URL valid for 10 minutes.

energent_get_status

ChatGPT
Get the current status of an audit started with energent_start_audit: running | needs_input | completed | failed | stopped, plus the latest task_id, a short summary and a few recent agent steps. When finished, read final_answer: the auditor's full reply (capped at 50 KB; final_answer_truncated says if it was cut); reports, if any, via energent_list_files. needs_input means the auditor asked a question or wants a plan approved; final_answer holds it, and energent_reply (or the web app at web_url) answers it.

energent_list_files

ChatGPT
List the files the audit's latest task produced (reports, corrected spreadsheets, charts). Pass a returned path to energent_get_file.

energent_prepare_upload

ChatGPT
For files too big to send inline (over 10 MB in total): returns one presigned HTTPS PUT URL per file. Without conversation_id it prepares the folder of a NEW audit and returns its conversation_id; with the conversation_id of an existing audit it uploads into that audit's folder (for energent_reaudit). PUT each file's bytes with the listed headers, then call energent_start_audit(conversation_id=..., uploaded_files=[...]) or energent_reaudit(..., uploaded_files=[...]) with the returned names. Uses one file upload from the plan's quota per file.

energent_reaudit

ChatGPT
Re-audit work after you fixed the findings of an earlier Energent audit. The new request continues the SAME conversation, so the auditor still has the earlier files, the data lineage it traced and its previous findings. Prefer this over a new energent_start_audit whenever it is the same piece of work. prompt is sent WORD FOR WORD. Say: what changed since the last audit (which findings you fixed and how), which attached files are the new versions, and exactly what to re-check (the fixed numbers or claims, plus anything the fixes could have affected). Attach updated files with files (inline base64, same limits as energent_start_audit) or uploaded_files (after energent_prepare_upload with this conversation_id); an existing file is never overwritten -- a new upload with the same name is stored as "name (2).ext" and the response lists the final names. Refused while the audit is still running or waiting for an answer: wait, reply or stop it first. Returns the new task_id; energent_wait, energent_list_files and energent_get_file then follow the re-audit.

energent_reply

ChatGPT
Answer an audit that is waiting for input (status needs_input: the auditor asked a question or wants its plan approved). content is sent word for word as the user's reply. If the audit is still running, the reply is queued and starts when the current task finishes (disposition 'queued'); if it is waiting, it starts at once (disposition 'started', with the new task_id). Refused for a completed, failed or stopped audit: use energent_reaudit to re-check updated work. Text only. Pass the same request_id when retrying so a reply is never sent twice.

energent_start_audit

ChatGPT
Start an Energent audit: an independent AI agent re-checks a piece of work (an analysis, a spreadsheet model, a report, a data pipeline's output) in its own sandbox and reports what holds up and what does not. It returns at once with a conversation_id; follow it with energent_wait / energent_get_status, then read the results with energent_list_files / energent_get_file. The user can also watch the audit live at web_url. prompt is sent to the auditor WORD FOR WORD as the first message, so write it as a complete, self-contained audit request. A good request states: 1. What the work is and who or what produced it (e.g. "a revenue forecast I built with pandas from the attached CSVs"). 2. Which attached files are SOURCES (inputs, raw data) and which are OUTPUTS to check (the deliverable). Name each file. 3. The specific numbers, claims or steps to verify (e.g. "the 2025 total of $4.2M in summary.xlsx!B12", "that duplicates were removed before the join"). 4. The form of answer wanted (e.g. "a pass/fail per claim with the corrected value, and a short report"). Do not include secrets. Attach files inline with files (base64, at most 20 files and 10 MB in total); nothing is fetched from URLs. For bigger files, call energent_prepare_upload first, PUT each file to its URL, then pass its conversation_id and the file names as uploaded_files. thinking_level trades speed for depth (default "high").

energent_stop

ChatGPT
Stop the audit's running task. Returns the honest outcome as stop_status: 'stopped' (it was running and is stopping), 'already_completed' (it had already finished) or 'not_running' (nothing was running). Work done so far stays in the conversation.

energent_wait

ChatGPT
Wait until an audit stops running (or timeout_s passes, max 50 s), sending progress notifications meanwhile, then return the same status as energent_get_status plus timed_out. Call it again while timed_out is true. When finished, read final_answer; reports, if any, via energent_list_files. Cancelling the wait never stops the audit.

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/energent-ai"><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/energent-ai)