MCP App Store
by KeenEthics
Breakthrough icon

Breakthrough MCP Server for ChatGPT

By Breakthrough

Overview

Connects your assistant to a living playbook distilled from your sales calls, so it drafts and answers with the context and the approach that actually works across your organization. Reads contacts, companies, call transcripts and your playbook from your own Breakthrough organization.

How to connect

  1. ChatGPT: Open the Breakthrough listing in the ChatGPT app directory and choose Connect. ChatGPT listing Breakthrough exposes 11 tools here.

Tools

business_profile

ChatGPT
What Breakthrough has on a company: description and past-call notes.

call_transcript_conversation

ChatGPT
Read one call transcript by id, in full or as a thematic summary.

contact_profile

ChatGPT
What Breakthrough has on a person: role, employer, background, and past-call notes.

contact_transcripts_list

ChatGPT
List the calls a contact was on (date, name, id), by contact id.

import_contact

ChatGPT
Add a new person to Breakthrough from their LinkedIn URL or email address. Side effects: - May enqueue create_contact_task (Celery) and trigger external enrichment. - Polls the DB for Contact/FitResult creation (up to ~90s by default).

refresh_contact

ChatGPT
Call for any reason a contact's record may be out of date: a direct request, a report that the person changed jobs or left, or a doubt about whether their details are current. Queues a re-pull of the existing contact's details to update their record on file; the update finishes after the tool returns started. Unless the user asked for the refresh outright, the tool returns needs_confirmation instead, so you can ask the user before anything is re-pulled. When the user TELLS you where someone works "Mauricio works at Autopistas del Café", "Carlo is at MO3 now", "she's moved to Odinsa" — pass that company as employer. Re-pulling LinkedIn cannot help here: the provider is the thing that was wrong, which is why this path exists. employer is where the person IS. A sentence can name a company for the opposite reason, and those two readings must never be confused: - "Venkat LEFT Splashtop", "she's no longer at Acme", "he's leaving Globex" name where the person is NOT. There is no stated employer here at all. Do NOT pass it as employer — that would file them at the company they just left, which is the exact corruption this tool has to avoid. Leave employer unset; this is an ordinary refresh, and if the user also says where they went ("left Splashtop for RudderStack"), the destination is the employer, never the origin. - A company inside a QUESTION — "is Matt still at Quindar?", "did Marco leave Algolia?" — is something to check, not a fact being supplied. Never pass it. Nothing is written until a person confirms. The tool walks one question at a time and each status names exactly what is still missing, so ASK THE USER and call again with the answer. Never fill these in yourself: - needs_contact_choice several people match. Show candidates and pass the chosen one's id as contact_id, keeping everything else the same. - needs_business_choice several companies match, or one partially matches. Show choices and pass the chosen employer_business_id. Do not pick for them. If none is right, ask for the website and pass it as employer_website. - needs_new_company_confirmed no company on file matches. If choices holds a company the tool found, ask whether it is the right one; otherwise ask for the website. Call again with employer unchanged (it is the company's name) and employer_website set to the confirmed web address. - needs_change_kind ask whether we simply had it wrong, or they have moved jobs. Pass employer_change_kind as "correction" or "job_change". This decides whether their PAST calls move too, so it must come from the user. - needs_since a job change needs a start date. Pass employer_since as YYYY-MM-DD. Calls before it stay with the old company. - proposal ready. Show message — it spells out what will change, including how many calls move — and, only once the user agrees, call again passing apply_token. - no_change_needed we already have them at that company and their calls are filed there. Say so; do not ask anything further. - applied done. - expired the confirmation lapsed or was already used; start over. apply_token executes the plan the server built and the user saw. It is single-use and short-lived, so do not store or reuse one, and never invent one. user_requested: set true only when the user asked for this refresh outright ("refresh Matt Wood", "re-pull her details", "update his record from LinkedIn") or has just confirmed a refresh you offered. Leave it false when you are acting on a hint that something changed ("Marco is now CEO of Kleecks", "is Matt still at Quindar?"); the tool then resolves the contact and returns needs_confirmation so you can ask before anything is re-pulled. The tool never infers this from the wording of query. Only a started status enqueues anything. Every other status is a safe no-op: - started refresh was queued; enrichment is NOT complete yet - needs_confirmation one contact resolved from an implicit signal — ask the user before refreshing, do not call again unconfirmed - ambiguous several candidates matched; pre…

resolve_prompt_context

ChatGPT
Call first for any prompt that could be about a person, company, or recorded call: finds the contacts, companies, and calls on record that the user's prompt refers to. Pass the prompt verbatim; it returns the matches with their ids and says which tool to call next, and an empty map when nothing on record matches. Call it before the other Breakthrough tools: it reads the whole prompt and tells them what is actually on record — which contacts, companies and calls match, with their ids and the filters that find them. The map tells you which specific entities exist in scope and includes concrete recommendations for the next tool call. Follow the recommendations unless you have a specific reason not to — they're generated from real Typesense counts against this org's data.

sales_methodology

ChatGPT
One company's sales-methodology scorecard (MEDDPICC or the org's own methodology): how the rep is running the deal against the team's process, what is established and what still needs digging into, from every recorded sales call. Internal and read-only; one company per call (it cannot list or compare deals). Use it when the request is about that deal: what to do next, gaps or blind spots, prep for a call with the company, strategy, or where the deal stands. business_profile shows a sales_methodology field when the company has a scorecard, and contact_profile shows company.sales_methodology for the person's company. It is not the notes of any one call: for "MEDDPICC notes from my last call", read that call. Each category shows its coverage, what is not yet covered, and the lead of the evidence (who, what was said, the risk). Pass categories with the category names shown in a previous result to get the full evidence for those; detail="full" gives it for every category (several thousand tokens). Keep scores out of anything the prospect reads. If you check several companies to answer a question across deals, say which ones you checked and that they may not be all of the user's deals.

sales_playbook

ChatGPT
Call for anything go-to-market: sections of the user's own company's sales and partner playbooks relevant to the user's prompt, the company's own strategy and know-how learned from its real deals and calls. Pass the latest user message verbatim as latest_user_message and the earlier turns as conversation_history so a short follow-up still carries its subject; it returns only the sections that apply and skipped when none do, so calling on anything borderline is fine and the answer reflects how this company operates rather than generic advice. This is the single highest-value tool for any "how should I approach this" or "write this for me" request. It is what makes the answer specific to this company rather than generic sales advice. Reach for it liberally: retrieval is cheap, it returns only the sections that match, and an empty result costs nothing. Call this whenever - The user asks how to handle, approach, position, pitch, frame, or sell something. - You are about to draft anything a prospect or partner will read — an email, a LinkedIn message, a proposal, a follow-up, a recap. Call it before writing, so the draft carries this org's positioning and language instead of generic copy. Do this even when the user did not mention the playbook. - The user asks what the company does, what makes it different, what its value or elevator pitch is, or how to explain it to a particular audience. - The user asks for discovery questions, talk tracks, next steps, or how to move a deal forward. - The user asks about a competitor, pricing, packaging, or a business case. - The user asks about partnerships, channel, or co-selling. Some orgs keep a separate partnership playbook, and this tool selects across all of them — so ask it for partner-motion questions exactly as you would for direct-sales ones. - The user names the playbook at all ("what's in my playbook", "using my playbook"). When in doubt, call it. The common failure is not calling it and producing advice that could have been written about any company. Do not use it for Facts about a specific person or company — use contact_profile / business_profile. What was actually said on a call — use search_transcripts. This tool returns org-level guidance, never per-entity data.

search_transcripts

ChatGPT
Search this company's call transcripts by topic, participant, business, or date.

whoami

ChatGPT
Who the signed-in user is and which organization this session reads from. Call this first in any task that will touch Breakthrough data — every other tool resolves against the same license scope, and a populated license_id is how you confirm the session is usable.

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/breakthrough"><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/breakthrough)