Build offering — discovery-optimized (USPs + philosophy)
Positioning of the build offer specifically — what sets a Ghost Team build apart. Companion to the pricing/delivery decision 2026-06-app-build-pricing-delivery and the discovery-niche ruling 2026-06-focus-discovery-niche (build is discovery-led, never a standalone agency play). Counter-position vs wani-wani’s open-source SDK (2026-07-03 call): full-service, sell the knowledge injection + optimization cycles + ranking focus — “build it yourself if you want, but you won’t rank number one.”
The USPs (near-verbatim — Vincent, founder chat 2026-07-03)
a) blazing fast servers b) reverse engineer user intents to optimize your discoverability c) optimized metadata per target registry (Claude, OpenAI) d) agent optimized integrations into your user conversion journeys (website handoff, etc) e) expert submission for high directory ranking f) staying up to date on the latest standards and building on the best tech (maintenance package)
The point that makes these BUILD USPs (not just discovery-service features): MCP servers are not ranked equally. [evidenced 2026-07-03] Some servers are demonstrably more optimized for discoverability than others — our ADO score spread across brands and the decoded registry mechanics show the same capability, packaged differently, ranks differently → ado-score-model, claude-registry-search-decoded.
Build philosophy — tools-first, not engineering-first
- START with the optimized tools you need to win the category, THEN engineer the capabilities. A product-market-fit framing for the customer rather than engineering-first — “common sense actually,” and a message product people immediately understand (Elliot, 2026-07-03: “would land super well”). Candidate killer slide.
- The diagnosis behind it —
[hypothesis]: “your API endpoints are not optimized for agent discoverability, so they should not be tools” — the biggest mistake companies make right now. Supporting datapoint [evidenced 2026-06-14]: 14% of ChatGPT apps fail to invoke on their own obvious demo prompts → product-thesis. - The crossing we’re uniquely positioned for: product-perspective performance × agent-perspective discoverability — we optimize both in build, and [evidenced 2026-07-04] the optimization engine is live (daily registry probes + lift advisor, 88% directional on 118 real edits → claude-registry-search-decoded).
Evidence status per USP (the honest map)
| USP | Where it stands |
|---|---|
| a) blazing fast servers | Build-quality feature. As a discoverability lever, latency is [hypothesis] (see the signal inventory → signal-set-and-weights) |
| b) reverse-engineer user intents | [evidenced 2026-06-24] prompt→keyword compression measured on ~4.3k runs → claude-connector-routing; the retrieval stage it feeds is [fact 2026-07-04] → claude-registry-search-decoded |
| c) optimized metadata per registry | Claude: [fact 2026-07-04] exactly 4 indexed fields, mechanics decoded → claude-registry-search-decoded. ChatGPT/OpenAI: [hypothesis] — architecture sprint under way (2026-07-03 call) |
| d) conversion-journey integrations (website handoff) | [idea] — no measured funnel evidence yet; the differentiator is the design capability |
| e) expert submission for high directory ranking | [hypothesis] — directory rank is refuted as a backend-retrieval signal but timing/freshness/novelty effects remain live candidates → signal-set-and-weights |
| f) standards/maintenance package | Service feature; underpinned by [hypothesis] that metadata degrades as models rotate → tool-description-optimisation |
Open questions
- Rank MCP servers publicly by how discoverability-optimized they are (Vincent, 2026-07-03: “a future thing we can investigate and rank”) — filed in ideas; natural lead-gen for the build offer.
- ChatGPT-side metadata mechanics (upgrades USP (c) beyond Claude) — next-sprint research focus.
Timeline
- [2026-07-03] (founder chat — Vincent + Elliot, recorded by Vincent) USPs a–f defined; build philosophy framed (tools-first: “start with what wins the category, then engineer the capabilities”; API endpoints ≠ tools). Elliot: message lands with product people; for the current deck, merged into a single “why us” slide (limited slides). Optimization engine noted live since 2026-07-02. Context: same-day Wani-SDK counter-positioning call (full-service vs open-source self-serve).
- [2026-07-07] (human-curated) Page created on Vincent’s instruction; USP bullets included near-verbatim; evidence-status map added per 2026-07-06-epistemic-status-standard.