How do I run it in Claude, ChatGPT, Codex and Gemini?
Web browser
Step 1: open ICP Deep Dive and describe your best customers (industry, size, contract value, sales cycle, trigger, champion). Step 2: open ICP Scoring Model, enter the traits they share with an importance for each, and press Run. For IT services or another services business, choose the business model services in ICP Deep Dive. No sign-in.
Claude
- Open Customize, then Connectors, select +, then Add custom connector.
- Name:
ICP Intelligence. Remote MCP server URL:https://icp-intelligence.gtmhelix.com/mcp. Leave the sign in fields empty and select Add. - Turn it on from the tools menu in a chat and ask in plain words.
Claude Code: claude mcp add --transport http icp-intelligence https://icp-intelligence.gtmhelix.com/mcp. Custom connectors work on every Claude plan; the Free plan allows one. Full Claude steps
ChatGPT
Install the Helix ICP Intelligence plugin from the ChatGPT plugin directory, then ask in a chat. It connects to the same ICP Intelligence tools with no sign in.
Or add a custom MCP server with https://icp-intelligence.gtmhelix.com/mcp and no authentication, if your workspace allows custom connections.
Codex
codex mcp add icp-intelligence --url https://icp-intelligence.gtmhelix.com/mcp
Then ask Codex in plain words.
Gemini CLI
gemini mcp add --scope user --transport http icp-intelligence https://icp-intelligence.gtmhelix.com/mcp
Or add all six Helix connectors and the plugin skills at once: gemini extensions install https://github.com/shashwatgtm/helix-gtm-suite
Then ask, for example
Use ICP Intelligence to build an ICP scoring model. Annual online payment volume is critical (above $50M, $10M to $50M, below $10M); payment gateways in use is important (3 or more, 2, 1); our best deals ran 3 or more gateways.
What does the output look like?
About this example. Juspay is a real company, named only to show how the tool works. Helix GTM Consulting has no relationship with it, and it has not reviewed this page. Its description comes from its own website. Every figure and detail in the inputs is an illustrative test input, not Juspay's data, plan or claim. Juspay's website calls it a global payments operating system and a payment orchestration platform (juspay.io).
Inputs used
- Company: Juspay. Product category: payment orchestration for online merchants.
- Four best deals, described by type only: an online travel agency, a quick commerce app, a D2C fashion retailer and a food delivery app, with gateways, payment volume, cycle, trigger and champion for each (all illustrative).
- Criteria for the scorecard: annual online payment volume (critical), payment gateways in use (important), merchant sector (important), checkout built in house (nice to have).
Output, cut from the real answer ([...] marks a cut)
## Synthesized ICP Based on pattern analysis (a value is named only where it leads; a tie lists every tied value): **Ideal Customer Profile**: - **Industry**: online travel, quick commerce, D2C retail and food delivery (tied: 4 industries equally common) - **Size**: 1,000 to 5,000 employees - **Budget**: $90,000 to $240,000 ACV seen in your customers (average $165,000) - **Tech Stack**: Uses own app checkout + 3 payment gateways - **Buying Trigger**: "a spike in failed payments during a sale", "moving checkout in house", "rising payment costs" and "an outage at one gateway" (tied) - **Champion**: Head of Payments - **Sales Cycle**: ~84 days (average of your deals) [...] ### Scoring Matrix | Criterion | Weight | Scoring Values | |-----------|--------|----------------| | **annual online payment volume** | 25 pts | above $50M (25 pts) / $10M to $50M (13 pts) / below $10M (0 pts) | | **payment gateways in use** | 15 pts | 3 or more (15 pts) / 2 (8 pts) / 1 (0 pts) | | **merchant sector** | 15 pts | online travel (15 pts) / food delivery and quick commerce (15 pts) / D2C retail (15 pts) / Anything not listed above (0 pts) | | **checkout built in house** | 10 pts | own app and web checkout (10 pts) / hosted checkout page (10 pts) / Anything not listed above (0 pts) | For merchant sector and checkout built in house, you listed the customers or roles you want, so every value you listed scores the full weight and a prospect that matches none of them scores 0. Every value in such a list scores the same, so the model does not rank your targets against each other yet: it tells a prospect inside your list from one outside it. If your closed deals show that one scores better, lower the others. For annual online payment volume and payment gateways in use, the values run in an order (a range, yes or no, a timeline), so the first value you listed scores the full weight, the last scores 0, and the values between score even steps; list each of these criteria's values from best fit to worst fit. All of these points are examples to adjust. **Maximum Score**: 65 points (Example figure: replace with your own) --- ## Qualification Tiers | Tier | Score Range | Action | SLA | |------|-------------|--------|-----| | **A: Hot** | 52-65 | Integration call and sandbox access | Sandbox access within 48 hours | | **B: Warm** | 39-51 | Priority follow-up with a volume and pricing discussion | Pricing call within a week | | **C: Developing** | 26-38 | Nurture with integration guides | Monthly touch | | **D: Unqualified** | 0-25 | Keep on the list; revisit when volume grows | Quarterly check | The bands are 80%, 60% and 40% of your maximum score of 65 points. Set the times above to how deals run in fintech, payments and banking APIs: Developer led evaluation in a sandbox, then a risk, compliance and commercial review; go live flow by flow or on a share of the traffic. ---
Real output from the ICP Intelligence connector, run on 19 November 2026. Your own inputs give a different answer.
What does it not do?
- The point weights are preset by importance level (critical 20 to 25, important 10 to 15, nice to have 5 to 10). The tool says so in its answer: they are not calculated from your deals.
- It does not pull deals from a CRM or score accounts until you give account evidence. Unscored criteria stay unscored, never 0 by default.
- ICP Deep Dive works only from the customers you describe and looks nothing up; a tie between traits is named as a tie.
- In Claude, ChatGPT, Codex or Gemini CLI (not the browser form) you can add results from deals you held back; if they miss your stop rule, the tool holds the scoring model instead of approving it.
Questions and answers
Can it build an ICP scoring model from my best deals?
Yes, in two steps. ICP Deep Dive finds what your best customers share; ICP Scoring Model turns the traits you choose into a points scorecard with A to D tiers. You decide which traits matter and how much.
Does it learn the weights from my data?
No. Weights are example points set by the importance you give each criterion, and the answer labels them as examples. Test them on deals you held back; through the connector in an AI app, the tool holds the model if that test fails.
Is there a version inside Claude?
Yes. Connect ICP Intelligence in Claude, or install the B2B ICP and Customer Insight plugin and use its lead scoring model skill, which scores leads and treats a missing value as incomplete, never 0.
Where next?
- ICP Scoring Model: the free web tool
- ICP Intelligence connector: all its tools and setup
- Bottom-up TAM, SAM, SOM and ICP scoring
- Lead scoring and MQL rules sales will trust
- Plugin: B2B ICP and Customer Insight (lead scoring model skill)
- All jobs: pick the job, run the tool
- All 89 tools and skills
- GTM Whisperer, the newsletter by Shashwat Ghosh
- Book 45 minutes with Shashwat when one tool is not enough