PitchBook answers “should I invest?” SalesBook answers “should I pilot?” — an MCP connector that lets AI assistants search 449 evidence-graded startups ranked by Fortune 500 sales-readiness.
Priority verticals: Retail & CPG, Manufacturing, Healthcare and Education, plus cross-industry platforms. We go deep in these rather than wide.
What it does
SalesBook ranks startups by buyer readiness, not fundability: how credibly a startup could sell into and deploy at a Fortune 500 company. Profiles are compiled from public sources with machine-assisted research, evidence-graded with sources, dates, and confidence labels, and scored against anonymized F500 buyer archetypes. Rankings are refreshed quarterly.
The connector exposes the registry over Streamable HTTP MCP — no API keys, no accounts, read-only. It works with Claude, ChatGPT, and any MCP-compatible assistant.
Add SalesBook to your AI assistant
Add this address as a connector. It is read-only, open to everyone, and needs no account or key.
https://mcp.maxeds.org/mcpStreamable HTTP · no sign-in
Claude
Open Settings → Connectors and choose Add custom connector.
Paste the URL above and choose Add.
Start a chat and ask, for example: “Using SalesBook, find AI startups that sell to hospital revenue-cycle teams.”
ChatGPT
Turn on developer mode under Settings → Apps → Advanced settings. On Business, Enterprise and Edu plans an admin may need to allow it first.
Under Settings → Apps, choose Create, enter the URL above, and choose Scan Tools. No authentication is needed.
Choose Create, then open a new chat and select SalesBook from the tools menu or mention it in your prompt.
Menu names change from time to time. Any assistant that supports remote MCP connectors can use the same URL.
Developers
Claude Code:
claude mcp add --transport http salesbook https://mcp.maxeds.org/mcp
To explore the tools by hand, open the MCP Inspector, choose the Streamable HTTP transport, and connect to the same URL.
Three prompts to start with
“Using SalesBook, which retail startups are strongest in shelf analytics and in-store operations for a large grocer? Show the top five with their scores.”
“Using SalesBook, find manufacturing startups for predictive maintenance, and show the deal tombstones behind the top two.”
“Using SalesBook, find AI startups that sell to hospital revenue-cycle teams, and tell me which ones have the thinnest evidence.”
The same works for education and for cross-industry (horizontal) platforms.
Tools
Five tools, all read-only.
search_startups Keyword search by name, problem or solution; the sector (retail, manufacturing, healthcare, education, horizontal) is optional.
get_profile The full evidence-graded profile for one startup, with each source and its date.
get_tombstones Dated, sourced public claims of enterprise wins, graded T1 (contract or rollout) to T4 (logo wall), claimed-unverified unless corroborated.
get_ranking The ranking for a sector and period (up to 100 per sector), with scores and the methodology version.
get_methodology The published weights, tier definitions, confidence labels and honesty rules.
More example prompts
“Show me the deal tombstones for Gatik — what enterprise wins are actually claimed, and how are they graded?”
“Rank the top 10 healthcare startups for an academic health system buyer, and explain what the score is made of.”
“What does SalesBook’s methodology weight most heavily, and how are tombstones graded from T1 to T4?”
Honesty notes
Sources. Profiles are compiled from public sources with machine-assisted research, not supplied by the startups themselves. Profiles are built on sourced evidence, with each source shown.
n=1 buyer archetypes. Rankings are scored against anonymized buyer archetypes (v0) modeled on real F500 buying patterns — a starting lens, not a universal buyer.
Claimed, not verified. Tombstones default to claimed-unverified. Founders can claim and correct profiles; corrections are marked founder-attested.
Thin-evidence tails. Lower-ranked profiles are flagged thinEvidence where public evidence is sparse.
The buyer ontology v0.1
The sectors in focus: Retail & CPG, Manufacturing, Healthcare, Education, and Horizontal (cross-industry platforms).
Rankings are scored against a structured buyer ontology — the consistent format for every evaluation. Fourteen fields, each labeled by how it is known:
Attested: honest differentiator, honest weakness, buyer-context fit (including honest non-fit), pilot data requirements, PoC speed.
Declared: the single categorized ask (design partner, pilot, paying customer, warm intros, capital).
Fields AI cannot verify carry low confidence and wait for founder correction. The ontology is v0.1 — derived from real buyer workflows, not universal yet.
Ranking methodology v0.2.7
Every profile carries two numbers. Readiness (0–100) is what ranks — “should I pilot?” Evidence grade (0–100) is displayed alongside and never mixed in — “how much should I trust it?”
Buyer fit ×30% — analyst score against the sector buying pattern.
Commercial proof ×25% — tombstones weighted by tier (T1=10, T2=5, T3=2, T4=0.5), confidence (corroborated=1.0, medium=0.75, claimed=0.5), and recency (fresh=1.0 down to 3yr+=0.5). Capped, then normalized.
Stage maturity ×15% — mean of stage badge (Early 30 / Growth 55 / Pilot-ready 70 / Established 80 / In-market 85 / Scaling 100) and a VC-backing signal: Series D+/growth ≥$100M = 100, Series C = 85, Series A/B = 70, Seed = 55, angel-only/undisclosed = 40. Unknown financing is excluded, never zeroed.
Pilot mechanics ×15% — data lightness, PoC speed, and a single categorized ask.
Team & trajectory ×10% — growth signal where published.
Evaluation completeness ×5% — the 12 buyer-ontology checks present.
Missing data rule: sparse proof/trajectory scores the corpus median and is labeled median-imputed — never zero, never silently reweighted. Other missing components are excluded with weights renormalized, and the exclusion is recorded. A profile ranks only with 3+ real signals, evidence grade ≥25, and institutional VC backing or a documented exception (profitability, bootstrapped scale, or strategic backing with evidence) — below any of these it is kept in the database unranked, with the reason published.
Education is our newest vertical: technology companies whose buyer is an institution — universities and colleges, community colleges, health-education programs, K-12 districts and schools, and state education departments. Consumer tutoring and test-prep, pure content publishers, companies acquired by a larger parent, and companies without a sourced venture round are left out. 49 companies qualified; we did not pad the list to 50.
Each company is scored 0–100 on four parts, and every fact links to a page opened on the date shown:
Institutional proof (0–40) — strongest when the institution itself, or independent news, names the adoption; lower for a company case study or a customer list.
Institutional scale (0–25) — the company’s stated count of institutions, districts or schools served.
Stage and recency (0–20) — the size and date of the latest venture round we could source.
Institution focus (0–15) — full marks for companies that sell almost only to education institutions.
Ties are broken by proof, then scale, then name. A profile is marked Sourced when website, description, proof, stage and headquarters (on the company’s own site) are all sourced, and Partial otherwise — most often because headquarters appears only on a third-party page. Scale counts are company-stated and sometimes dated; proof grades the public claim, not customer satisfaction.
A profile is ranked only if it meets all of these. Otherwise it is kept in the registry, unranked, with the reason published — never silently dropped.
Include: active, standalone startup (not bankrupt, merged, acquired, or a product line); VC-backed or a documented exception; 3+ readiness signals with data; evidence grade ≥25; operating in-market.
Exclude → Historical record: terminal legal events — bankruptcy, merger, acquisition. Kept with evidence, no rank.
Exclude → Status watch: credible reported distress (layoffs, valuation collapse, liability management, founder exits, viability-threatening lawsuits). Claims are always source-attributed (“Per <source>: <claim>”), never asserted in our voice. Visible and searchable with per-source notes; no rank number; returns to ranking only when the reporting is resolved.
Exclude → unranked record: no institutional backing and no documented exception; or below the evidence bar (“thin evidence — unranked pending verification”).
Current Status watch lane (Oct 5 2026): Innovaccer, FinThrive (healthcare); Cooler Screens, Covariant (retail).