Manufact

Canonical name: Manufact — https://manufact.com/. Alias / mishearing: transcripts/Granola render it as “Manufactured” / “Manufacture” / “manufacture” / “Manfuck” / “manicure” — all map to Manufact. An MCP hosting/SDK platform (like alpic): SDK + open-source framework + hosting; also has a website-chat SDK and a Vibe-Code-style app builder (reportedly weak per han/secondhand).

Current truth

• Dual role: potential customer + partnership channel. Track Manufact as a customer because their own connector can use Ghost Team discoverability data/optimization, and as a partnership because their customer base can create leads and a learning loop for what MCP builders actually want.
• Preferred hosting partner — stronger mutual fit than alpic. Manufact is not planning enterprise/connector builds (explicitly told us) and is actively looking for a build partner — which is us. So the synergy is mutual, where with Alpic we’re the weaker hand.
• Three partnership angles (agreed in principle, 2026-06-19; debriefed 2026-06-22):
  1. Ghost Team → Manufact: we refer hosting leads to them.
  2. Manufact → Ghost Team: they refer connector builds to us.
  3. Ghost Team → Manufact: we share organic discoverability data for Manufact and their customers; they can show a branded Ghost Team discovery score per server on their platform.
• Lead/customer-learning loop: Manufact already shares agentdiscoverability.com / Ghost Team tooling with customers organically. Their customers split into useful learning segments: B2B/developer-tool companies with real Claude-connector motivation, and larger European/B2C companies that mainly want board-level “AI first” presence.
• Anthropic relationship strengthening: Anthropic is fast-tracking Manufact’s customer submissions — a good durability signal. They’re seeing strong B2B / DevTools traction through Claude connectors themselves.
• Integration ideas to develop: the Manufact API could push Ghost Team recommendations as PRs or auto-applied changes to a customer’s MCP server; Manufact’s cloud analytics could expose per-tool error/usage data for Ghost Team to match against organic prompt tracking. Together this could close the loop from prompt visibility → recommendation → live server change → usage/quality evidence.
• Strategy: partner with both, non-exclusive. Candidate scoping: limit a Manufact partnership to Claude connectors, keeping the ChatGPT door open for Alpic (where Alpic is stronger). Vincent floated that real trust may need some exclusivity; left open.
• Durability caveat: need to know their raise/runway before building a company on them — “if they pull the plug in 18 months it’s annoying.” Alpic is the safer-runway bet (≥4 years).
• Game-theory risk (same as Alpic): Manufact could build a discovery engine themselves — every hosting platform is incentivized to add a “discovery tab” to keep apps on-platform. Our moat: a vast market (any company, incl. the many not hosted on these platforms), a multi-week-to-build organic-discovery capability (not a feature you toggle on), and more data. We must be the embedded discovery layer before platforms add their own.

Open questions

• How much did Manufact raise / what’s the runway? (Gates how much to build on them.)
• Does their API actually support programmatic resubmission / automated submission (the “close-the-loop” mechanism)? — Vincent to verify by trying the platform as a user.
• Which Manufact customers are good first lead/learning candidates for Ghost Team discovery data?
• Exclusivity: none, Claude-only scope, or some mutual commitment? Resolve as the partnership firms up.
• What is Anthropic’s Claude-connector popularity scale (the published ~5,000–25,000 range per connector) actually measuring? pietro-zullo to ask the Anthropic MCP team.

Contacts (2026-06-24): elliot-garreffa owns the Manufact relationship on the Ghost Team side (Pietro reached out to Elliot; the 2026-06-19 call was Elliot’s). Their-side contact is the founder pietro-zullo (pietro@manufact.com), likely also the unnamed “Them” on 2026-06-19 though unconfirmed.

Timeline

• [2026-06-24] (call — Vincent ↔ pietro-zullo, founder) First direct Vincent↔Pietro catch-up since the April MCP Dev Summit (NY). Manufact progress: hired two engineers in SF; positions as “best place to deploy MCP servers,” Vercel-for-MCP, developer-focused (YC companies + top dev tools as customers); features now include hosted gateway, in-app chat to test, tool/resource tests, agent-driven evals (runs your MCP via an agent / ChatGPT), publishing/submission-readiness checks, auto-generated submission assets, and per-branch review deployments with a public chat. Vibe (no-code MCP builder) quietly deprioritised/hidden — onboarding now hands you a prompt to paste into your own coding agent. Manufact uses Aura for agent-readiness and ranks ~3rd in the dev-tool category (on Aura’s landing page). Confirmed two customer segments: dev-tool startups = tens of thousands of tool calls/day (real traction, closer to cloud-connector/code); big (often European) consumer companies = low volume, board-level “AI-first” gimmick. Their own Claude connector has trended on Claude 2+ weeks, driving a big signup/connection surge — new apps get a “new”-tag boost to the top of the ranking then get deranked over time (logic undisclosed, Pietro’s read: tied to tool errors + other metrics). Analytics: Manufact logs every tool call (client identity, geography, usage); ChatGPT sends a unique user ID on the initialized request, Claude does not; geography is derived from client header metadata, standardised across hosts; Pietro rates Claude’s own analytics dashboard highly and is redesigning Manufact’s tab to mirror it. Concrete offers/commitments: generous platform discounts + a direct feature-request Slack channel for us; a possible analytics API endpoint for Ghost Team to consume directly (mechanism for partnership angle 3); referral of inbound MCP-build leads (esp. Italian companies — they don’t do builds); to ask Anthropic about the Claude popularity scale; to share anonymised Claude approval-time data. Claude approval process: still primarily manual, “effectively a random variable” (Pietro’s apps: 2 days–3 weeks; Anthropic close to auto-submission); some categories gated pending legal (e.g. insurance — “legal is all over the place”); Manufact’s own numbers would be biased since they brute-force submissions. Shared interest: an open-source, crowdsourced connector approval-time tracker combining Ghost Team’s ChatGPT data (~1.5-day avg) with Manufact’s Claude submission data. Vincent to trial Manufact for internal research/MCP experiments when bandwidth allows.
• [2026-06-22] (internal debrief — Elliot to Vincent) Replayed the 2026-06-19 Manufact conversation in the wider website/product/partnership strategy session. Strategic stance set: partner with both Manufact and alpic, non-exclusive, possibly scope Manufact to Claude connectors. See partnerships, alpic, and 2026-06-15-offering-pricing-fundraise.
• [2026-06-19] (call — Elliot ↔ Manufact; raw transcript title mishears as “Manufacture”) Good partnership/customer conversation. Manufact said Claude connectors are showing strong B2B/devtools traction; some YC companies reportedly close ~$30k/day through Claude connectors; Anthropic is helping fast-track Manufact customer submissions. Agreed three collaboration angles: Ghost Team refers hosting leads to Manufact; Manufact refers connector-build leads to Ghost Team because they do not plan to do enterprise builds; Ghost Team shares organic discoverability data for Manufact and relevant customers. Manufact already shares agentdiscoverability.com with customers. Integration ideas: Manufact dashboard could surface Ghost Team discovery score per server; Manufact API could let Ghost Team recommendations become PRs or auto-applied server changes; Manufact cloud analytics/per-tool error data could feed Ghost Team’s usage-vs-discovery matching. Next relationship moves: set up a shared Slack channel, have Vincent test Manufact’s platform/SDK, and share Manufact-specific organic discovery data.