MCP App Store
by KeenEthics
Konshus icon

Konshus MCP Server for ChatGPT

By Konshus

Overview

Konshus is a private memory you own. Bring in your history once — past AI chats, notes, journals, documents and 20+ other sources — and Konshus turns it into memories that each show where they came from. Once connected, your assistant can read a short summary of who you are before replying, search your memories and original sources, look back to what you thought at an earlier date, see what has changed since a date, and save a new memory when you ask. Every tool reads only the signed-in member's own data. You can export or erase everything at any time.

How to connect

  1. ChatGPT: Open the Konshus listing in the ChatGPT app directory and choose Connect. ChatGPT listing Konshus exposes 18 tools here.

Tools

check_upload_status

ChatGPT
Call BEFORE telling the member you can't find or open a file they uploaded to Konshus. Looks up their uploads by part of the file name and returns the true status (readable, still being read, too big, unsupported) with one plain sentence to tell them.

explain_what_you_know_about

ChatGPT
Prefer get_member_context (with depth/focus/since). Ask the member's Konshus to explain what it knows about a topic, person, or theme — returns a synthesized paragraph WITH inline citations back to source memories. Use for: 'what does Konshus know about my mom / my business / my health?' or when the caller wants a coherent narrative instead of a raw atom list. Combines semantic recall + short-form synthesis. Cite atoms by number so the member can trust the answer. Falls back to a raw atom list if synthesis fails.

get_artifact

ChatGPT
Read the FULL text of one source document in the member's Vault — a PDF or Word doc they uploaded, a whole conversation export, a Notion page, a journal entry. Use this whenever the member asks you to reference, quote, summarise, review or work from a specific document. WORKFLOW: call search_artifacts first to find the document and get its id, then call get_artifact with that id BEFORE saying anything about its contents. Never claim you can't see a document's text until you have actually called this. Long documents are paginated — the response tells you page 2 of 9; call again with the next page to keep reading. Everything in the Vault is readable this way EXCEPT files the member marked Private, which return a clear explanation instead.

get_big_picture

ChatGPT
Also included in get_member_context depth="full". Return the member's confirmed big-picture notes: short summaries of the arcs and themes running through their life that single facts miss (each one the member read and kept). Call when the member asks for the big picture, what's going on in their life lately, or broad 'who am I' questions.

get_briefing

ChatGPT
Prefer get_member_context (with depth/focus/since). Load the member's 1–2 page Briefing — deeper than Whisper, covering values, decision patterns, current focus, and recent context. Use when the conversation calls for richer grounding (planning, advice, multi-turn discussions, anything where Whisper feels too thin).

get_member_context

ChatGPT
The member's Konshus profile: who they are, how they want answers, and what they're working on, with the date it was built. Call it once at the start of a conversation, before your first reply, and again if they ask what you know or remember about them. depth="full" sends more lines in the same format; focus puts one topic first; area ("work", "personal" or a project) shows only that part; since (YYYY-MM-DD) shows only what changed. For a specific fact, use search_atoms.

get_persona_summary

ChatGPT
The member's full Konshus record: the same profile as get_member_context, same format and tags, with every active memory and its id. Use when the short profile isn't enough.

get_recent_knowledge

ChatGPT
Brain, not Voice. Returns the member's recent Knowledge lens atoms — reference material (docs, research, articles, notes) they've saved into the vault. Use when you want to know what topics the member has been actively studying or loading. Doesn't reflect who they are — for identity/voice use get_whisper or get_briefing. Grouped by source artifact so you can see clusters (e.g. 'Hawaiian sovereignty — 6 atoms from 2 docs').

get_vault_stats

ChatGPT
Return the member's real Vault inventory: total atoms (distilled insights), total raw artifacts (imported items like ChatGPT conversations, PDFs, Readwise highlights, Obsidian notes, etc.) broken down by source, and the date range of their data. Call this when the member asks 'how much is in my Vault', 'what do you have on me', or anything about Vault size/contents. Don't answer from memory — always call this.

ping

ChatGPT
Health check. Returns 'pong'. Used by the Konshus status page.

read_source

ChatGPT
Read the FULL text of one source document in the member's Vault — a PDF or Word doc they uploaded, a whole conversation export, a Notion page, a journal entry. Use this whenever the member asks you to reference, quote, summarise, review or work from a specific document. WORKFLOW: call search_artifacts first to find the document and get its id, then call get_artifact with that id BEFORE saying anything about its contents. Never claim you can't see a document's text until you have actually called this. Long documents are paginated — the response tells you page 2 of 9; call again with the next page to keep reading. Everything in the Vault is readable this way EXCEPT files the member marked Private, which return a clear explanation instead.

recall_past_self

ChatGPT
Date-anchored recall — what the member was thinking/doing around a specific past date. Returns relevant atoms from that window. Needs 90+ days of history and 100+ memories; otherwise it says so plainly.

save_chat

ChatGPT
Save a short summary of this conversation to the member's Konshus Vault after they ask (e.g. 'save this chat', 'remember this conversation'). Send a brief summary and, optionally, a few of the member's own key lines quoted word for word. Don't send the full conversation. Returns a receipt to show the member.

save_memory

ChatGPT
Save lasting facts, preferences or decisions the member shared, in plain words, written as "You …". Use it when something would change future answers: a decision, a correction, a new preference, a person or project that matters. Collect what's worth keeping and save it together: pass several facts in items (up to 10) in ONE call instead of many calls. Unless the member has turned on auto-save, ask once at a natural pause ("Want me to save these 3?") and save only after they agree — never ask after every message. After saving, show the member the receipt this tool returns. Never say something was saved unless this call succeeded. For a price, launch or decision, name the product or project in the text ("Konshus Plus costs…", not "It costs…").

search_artifacts

ChatGPT
Search the member's raw imported memory — also the tool for upload questions: when the member asks 'did my file come in?', 'did my upload work?' or 'can you see the PDF I sent?', call this (source_type document_upload, or the source they name) and answer from the newest result's title and date — don't answer upload questions from get_member_context alone. Covers ChatGPT/Claude/Gemini/Grok conversations, Obsidian notes, Readwise highlights, Day One/journal entries, Notion pages, uploaded PDFs/docs, Limitless transcripts, social posts, calendar events. Layer beneath search_atoms, usually much larger (thousands of items). Note: this returns a short excerpt, not the document. Never quote, summarise or characterise a document's contents from an excerpt — call get_artifact with the result's id first and read the actual text. Use when (a) search_atoms came up empty, or (b) the member names a specific source. When the member refers to a file they uploaded — 'my patent', 'the contract', 'that PDF', 'my resume' — pass source_type: document_upload on the first attempt; that is where uploaded PDFs and Word docs live and it makes the scan far more accurate. If the first query returns nothing, try again with fewer, more distinctive words (search on 'patent', not 'my provisional patent filing') before telling the member it isn't there. When the member names a source ('my PDFs', 'what I told ChatGPT', 'my Readwise highlights', 'my Obsidian notes', 'my journal', 'my Notion pages', 'my Limitless transcripts'), always pass source_type so the scan is narrow — never search the whole corpus when the member already told you where to look. Source map: PDF/doc→document_upload, ChatGPT→chatgpt_conversation, Claude→claude_conversation, Gemini→gemini_conversation, Grok→grok_conversation, Readwise→readwise_highlight, Obsidian→obsidian_note, Notion→notion_page, journal→journal_entry (also day_one_entry), Limitless/pendant→limitless_lifelog, Facebook/Instagram/LinkedIn→social_post, calendar→calendar_event. Returns title, source, date, excerpt, vault link.

search_memory

ChatGPT
Search the member's confirmed memories for one fact. Cite dates and confirmation state; summaries are not source quotations (use get_artifact). Call get_member_context before your first reply. For named source files use search_artifacts. If no useful match, call search_artifacts once; never loop with synonyms. For a cancelled project's earlier decisions use include_history=true with its exact project name as query: those results are explicitly set aside, never current facts.

search_sources

ChatGPT
Search the member's raw imported memory — also the tool for upload questions: when the member asks 'did my file come in?', 'did my upload work?' or 'can you see the PDF I sent?', call this (source_type document_upload, or the source they name) and answer from the newest result's title and date — don't answer upload questions from get_member_context alone. Covers ChatGPT/Claude/Gemini/Grok conversations, Obsidian notes, Readwise highlights, Day One/journal entries, Notion pages, uploaded PDFs/docs, Limitless transcripts, social posts, calendar events. Layer beneath search_atoms, usually much larger (thousands of items). Note: this returns a short excerpt, not the document. Never quote, summarise or characterise a document's contents from an excerpt — call get_artifact with the result's id first and read the actual text. Use when (a) search_atoms came up empty, or (b) the member names a specific source. When the member refers to a file they uploaded — 'my patent', 'the contract', 'that PDF', 'my resume' — pass source_type: document_upload on the first attempt; that is where uploaded PDFs and Word docs live and it makes the scan far more accurate. If the first query returns nothing, try again with fewer, more distinctive words (search on 'patent', not 'my provisional patent filing') before telling the member it isn't there. When the member names a source ('my PDFs', 'what I told ChatGPT', 'my Readwise highlights', 'my Obsidian notes', 'my journal', 'my Notion pages', 'my Limitless transcripts'), always pass source_type so the scan is narrow — never search the whole corpus when the member already told you where to look. Source map: PDF/doc→document_upload, ChatGPT→chatgpt_conversation, Claude→claude_conversation, Gemini→gemini_conversation, Grok→grok_conversation, Readwise→readwise_highlight, Obsidian→obsidian_note, Notion→notion_page, journal→journal_entry (also day_one_entry), Limitless/pendant→limitless_lifelog, Facebook/Instagram/LinkedIn→social_post, calendar→calendar_event. Returns title, source, date, excerpt, vault link.

whats_changed_since

ChatGPT
Prefer get_member_context (with depth/focus/since). Standup / catch-up primitive. Returns the memories added or edited since a given date. Use for: 'what have I been thinking about this week?', 'catch me up since Monday', daily/weekly review workflows. Groups by day and includes atom type. Prefer this over search_atoms when the caller wants recency, not topical relevance.

Capabilities

Writes

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/konshus"><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/konshus)