Ideal Customer Profile MCP

An ideal customer profile you can talk to.

Derived from your closed-won deals, with a receipt on every number, and it tells you where it will not answer. Connect Telepath to Claude and ask your pipeline questions in plain English.

Claude today. ChatGPT, Gemini and the connector directories follow OAuth support.

What a conversation looks like

What's in my pipeline this week, by group?
→ get_quadrant_summary

Your open book is 78 deals. By action group: close 7 (total amount 201,500), rescue 20 (684,000), watch 15 (762,500), drop 25 (1,364,500). Every deal in every group carries an amount, so the totals rest on all of them. Amounts are in your HubSpot portal currency, which Telepath does not capture, so I am not attaching a symbol.

11 deals are unclassified: they have no published T-Score or W-Score, so they sit outside all four groups. They add up to 475,500. 7 + 20 + 15 + 25 + 11 = 78, for this response only.

16 deals sit within 5 points of a group edge. The nearest is deal 511983729874, in close, 2 points from the edge; across it lies watch. That is arithmetic on today's published scores, not a forecast of movement.

Population claim book:4cf6ec9630cef3b853a3e6854b47ae12 · scored 21 Sept 2026, 02:00 UTC · model d0d73988…
Why did [Company A] score 62?
→ explain_score

[Company A] — RevOps Rollout scored 62, segment Warm, confidence Medium (79% weighted completeness, no fields missing).

Where the raw fit came from, by share of contribution: industry 44.7% (fit 100%), overall resemblance to won deals (vector similarity) 22.3% (fit 62%), geography 15% (fit 100%), decision-maker level 8.2% (fit 100%), company size 7.1% (fit 100%), decision-maker function 1.7% (fit 20%), amount 1% (fit 100%). stage fit 0% and contributed nothing.

How the raw fit became 62: the raw fit is 0.82776276950135. Its nearest segment is segment 4, whose held-out won-deal fits (13 of them, from 0.5416885717280183 to 0.9857868462214914) place it at percentile rank 0.6153846153846154. 100 × that, rounded, is 62, which is the recorded score. Model version 13.

Claim e29ec39e… · scored 18 Sept 2026, 16:43 UTC · model v13
Which deals can't you score, and what's missing?
→ get_abstentions

Telepath declines to score these (first 3, more on the next page): [Company B] — Annual (Contract Sent, completeness 16%); [Company C] — RevOps Rollout (Presentation Scheduled, completeness 28%); [Company D] — Annual (Decision Maker Bought-In, completeness 16%). The reason on each is facet:tier-1-completeness: too little of the profile is filled in for an honest score, so the answer is a dash, not a number.

What is missing on [Company B] — Annual: Company Size, Region, Decision Maker Seniority, Industry, Tech Stack, Lead Source, Decision Maker Department, Sales Cycle, Stakeholder Count. Fill those in HubSpot and the next run can score it.

Last run 21 Sept 2026, 02:00 UTC · model d0d73988…
Can you check that 62 is real?
→ verify_claim

Attested. Telepath emitted this claim. Checked against the append-only score ledger, by id: a T-Score event, score 62, segment Warm, for deal 511983729874, scored 18 Sept 2026, 16:43 UTC under model v13.

Claim e29ec39e-fb70-4e96-a536-80cf0bf872a1
Will the [Company A] deal close?

I can't tell you that, and Telepath won't either. Telepath is built from the deals your workspace won; nothing it returns is a probability of closing, a forecast, or a prediction of what a deal will do.

What it does hold on [Company A]: how closely the deal resembles your won deals (T-Score 62, segment Warm), the action group that puts it in (close, 2 points from the edge), and that no fields are missing. Ask me for any of those with a receipt.

No tool call. Narration Contract v0.2, rule 11: no forward projection.

Every number above came from a Telepath tool call and carries a claim id. Claude narrates; it doesn't compute. Ask it to verify anything.

Generated from a real workspace on 2026-09-21 with company, deal and rep names replaced. The figures move every night at 02:00 UTC; the shape does not.

What it won’t do

Will it forecast?
No. It won't tell you whether a deal will close. Telepath is built from the deals you won; predicting the future from that alone is guessing, and it refuses to guess.
Will it invent a total?
No. Telepath marks which numbers it computed, and Claude is instructed to say a figure Telepath did not compute cannot be attested rather than adding rows up itself. Ask it to verify any number and it will check the ledger.
Will it write to your CRM?
No. Read-only, enforced at the database, not in the prompt.
Will it score a deal you describe to it?
No. Scores come from your synced data through the ledger, never from a chat message.
Will it tell you what to say on a call?
No. It describes; it doesn't instruct.
Will it rank your reps?
No. No per-owner aggregation, by design.

What you can ask

Eleven tools, all read-only. Claude picks the tool; you ask in plain English.

  • verify_claimChecks a claim id, a model version id or a population id against the ledger, or a quoted score against every event in your workspace.
  • get_quadrant_summaryThe whole open book per action group: count, total amount with how many deals it rests on, the deals near a group edge, the unclassified residual, and a population claim id.
  • get_deal_scoreA deal's latest T-Score and W-Score from the ledger, with its action group and receipt. Null is an abstention, never zero.
  • get_company_scoreA company's entity score with coverage and the fields that would unlock it. Below the floor, the dash is the answer.
  • explain_scoreThe criteria and weights behind a score, and the final step: raw fit, nearest segment, held-out distribution, percentile, and the score reconstructed from them.
  • get_abstentionsWhat Telepath withheld and why, per deal or company, with the missing fields, when it last ran and under which model.
  • get_stale_scoresOpen deals whose score was made under a model other than the current one, naming both versions. A filter, not a recomputation.
  • get_driftWhether the won deals your model was built from have moved since it was built: fresh, stale, or an abstention with the reason.
  • get_icp_profileYour profile: engine figures anchored to a model version, authored text marked as authored.
  • list_by_quadrantOne action group's deals, paged, with the population count and the unclassified residual beside it.
  • get_signalsEnrichment citations for a company: source, tier, when retrieved, and the accepted value, fenced as untrusted text.

How it’s built to be trusted

  • Read-only: a SELECT-only database role. There is no write path.
  • Tenant-isolated, with a test that fails the build if it isn't.
  • Rate-limited. Test keys refused.
  • Every payload carries a claim id, the time it was scored, and the model version. verify_claim checks any of them against the ledger.
  • Your key can be revoked and re-issued at any time; a revoked key is refused on every tool from that moment.

Connect in two minutes

  1. In Claude: Settings → Connectors → Add custom connector.
  2. URL: https://api.telepath.pro/api/mcp
  3. Authentication: None. Additional header: x-api-key → paste the key Telepath issued you.
  4. Save. Ask: "What's in my pipeline this week, by group?"
  5. If it refuses something, that's working as designed. Ask it why.

We issue keys by hand during the founding period: once your subscription is active, Tom sends yours, and revokes or rotates it on request the same day.

What’s coming

  • What changed since last week (get_pipeline_changes). In build.
  • Which won deals this most resembles (get_similar_won_deals). In build.
  • Closed outcomes by fit band (how many won, how many lost, and cycle time), when there are enough to say. Telepath only started scoring deals while open in July 2026; it will tell you how many outcomes it has and how many it needs.

Included with every Telepath plan.

See pricing →

Not ready to connect HubSpot? Start with the free ICP report. Get the free report →