Proprietary corpus · MCP-native · evidence-cited

The competitive landscape
before you wire the check.

MarketPrior generates competitive landscape reports, distribution channel maps, and advertising opportunity briefs from a proprietary corpus — SEC EDGAR + Crunchbase open data + verified press feeds. The intelligence CB Insights charges $60K for, built for founders at founder prices. Every claim cites a corpus row; Pro re-verifies each source.

See a sample report How the 7-axis intake works →

7intake axesstructured input for the report query

~6page reportcompetitors · channels · founder fit

100%cited claimsevery fact links to a corpus row

cursor · mcp · scale4

> scale4.start({
  pitch: "AI-native contract review for mid-market legal teams"
})

// 200 OK · session opened · signed
{
  session: "srv1.eyJ2IjoiMSIsInNpZ...",
  protocol: "7-axis@2026.06",
  axes:    7,
  next:    "collect signals — submit_signal per axis. score gated until all minimum signals in."
}

scale4.submit_signal({
  axis:     "problem",
  kind:     "user_quote",
  value:    { verbatim: "We spend 40h/week on first-pass contract review..." },
  citation: "https://news.ycombinator.com/..."
})

// Accepted: problem.user_quote. 2/3 required.✓ signal validated · token refreshed

The structural failure

The model has no commitment to grounding.

Nothing in the default agent loop forces it to find a real user quote, name a real competitor, or explain why the incumbents haven't already done this. It can simulate validation as easily as it can do validation.

Sycophancy

"Great idea — here's the roadmap."

The model cheers, outlines a sprint plan, names a tech stack. No user quote required. No competitor named. Six months of work for an idea that fell apart on the first customer call.

Template recycling

Generic SWOT, zero citations.

Four-paragraph regurgitation of every business-school template the model has ever seen. Reads sophisticated. Nothing in it is falsifiable. You've learned nothing you didn't already know.

Hallucinated research

Five competitors, three of which don't exist.

Confident lists of named companies, half of which are miscategorised or fabricated. The agent simulates market research as convincingly as it does market research, and you can't tell the difference.

MarketPrior fixes this with a state machine that refuses to return a score until evidence has been collected — verbatim quotes, cited URLs, named competitors, one counterfactual for every axis. The protocol is the locked door the agent can't walk around.

The 7-axis protocol

No score without evidence.

Each axis declares its minimum signal set. The agent collects those signals using its own tools — web search, the user's clarifying answers, pasted links — and submits them back. The server checks shape and citations; the agent does the semantic work.

Axis 1

Problem hypothesis

Confirm the pain is real and held by specific people.

Required signals

user_quote ×3 (verbatim, URL, date, speaker role) · counterfactual_quote ×1 (someone for whom the pain is NOT acute, with explanation)

Axis 2

Beachhead market

Identify the smallest market you can win completely.

Required signals

icp_definition ×1 (named segment) · named_prospect ×3 (org + URL) · acv_estimate ×1 (with comparable-pricing citation)

Axis 3

Existing solutions

Map the competitive landscape with real pricing and real dissatisfaction.

Required signals

named_competitor ×5 (URL required) · dissatisfaction_quote ×3 (per-competitor) · pricing_row ×3 (comparable pricing table)

Axis 4

Wedge

Define the single entry vector that incumbents cannot quickly copy.

Required signals

wedge_statement ×1 (3-sentence max) · anti_wedge ×1 (why incumbents haven't done this — specific reason required) · replication_estimate ×1 (time-to-copy estimate)

Axis 5

Distribution

Name one channel and prove the unit economics of acquiring through it.

Required signals

channel_choice ×1 (specific, not "content marketing") · cac_benchmark ×1 (comparable cost data, cited) · founder_channel_fit ×1 (why this founder in this channel)

Axis 6

Unit economics

Confirm the business has a path to positive margin before the runway ends.

Required signals

cac ×1 (with sourcing) · ltv ×1 (with churn assumption) · gross_margin ×1 · breakeven_customers ×1

Axis 7

Founder fit

Confirm the founder has earned the right to win this specific market.

Required signals

domain_years ×1 (specifics, not "I've always been interested in X") · unfair_advantage ×1 (structural, not motivational) · shipped_artifact ×1 (public link to prior work)

Scoring

Maximum 70 points. Verdict flows from the math.

Every axis scores 0–10. The total determines the verdict; kill thresholds fire on individual axis weakness before the total even matters. A single axis at 3 or below triggers kill-or-pivot regardless of how well the rest scored.

buildall ≥ 7 AND total ≥ 56

iteratetotal 42–55, no axis ≤ 4

scope cutthree or more axes ≤ 5

pivotany single axis ≤ 3

killpivot + founder fit ≤ 3

The protocol flow

A state machine the agent walks. Not a prompt.

The server enforces ordering on tool calls and refuses to advance until evidence is in. That's the structural difference between asking an LLM nicely and wiring a locked door.

Agent calls scale4.start

The pitch text opens a session. The server returns the 7-axis protocol spec and a signed session token. No storage — the token travels with the agent.

Agent collects signals, axis by axis

Each submit_signal call requires a citation URL. The server HEAD-checks it. Placeholder text and generic quotes are blocked by regex validators. Evidence is non-negotiable.

score() unlocks when all axes are complete

The server refuses to score until every axis has its minimum signal count. Returns a 70-point scorecard, per-axis scores, and a verdict computed from the thresholds.

Roadmap + handoff emitted

generate_roadmap turns the scored axes into risk-weighted milestones. A BUILD verdict hands off directly to Stackforge's compose_stack. KILL emits the evidence ledger.

Scoring rubric — problem axis

The disconfirming signal is required by design.

Every axis carries at least one signal that requires evidence against the idea. You cannot score problem hypothesis without a counterfactual quote. You cannot score solutions without naming five competitors. The rubric makes cherry-picking structurally impossible — the score drops to 3 if you skip it.

8 pts≥5 user quotes + ≥1 counterfactual

6 pts≥3 user quotes + ≥1 counterfactual

3 pts≥3 user quotes, no counterfactual

0 ptsfewer than 3 user quotes

{
  "id": "problem",
  "version": "2026.06.01",
  "required_signals": [
    {
      "kind": "user_quote",
      "min_count": 3,
      "schema": {
        "verbatim":     "string, min 60 chars",
        "source_url":   "url, HEAD-checked",
        "source_date":  "iso8601",
        "speaker_role": "string, min 6 chars"
      }
    },
    {
      "kind": "counterfactual_quote",
      "min_count": 1,
      "schema": {
        "verbatim":    "string, min 40 chars",
        "source_url":  "url, HEAD-checked",
        "explanation": "string, min 100 chars"
      }
    }
  ],
  "scoring": {
    "rubric": [
      { "if": "quotes>=5 && counter>=1", "pts": 8 },
      { "if": "quotes>=3 && counter>=1", "pts": 6 },
      { "if": "quotes>=3 && counter==0", "pts": 3 },
      { "if": "quotes<3",               "pts": 0 }
    ]
  },
  "signature": { "algo": "ed25519", ... }
}

The five verdicts

The recommendation flows from the math.

MarketPrior does not have opinions. It has thresholds. The verdict is computed from the scores; the scores are computed from the evidence submitted. The tool can't be friendlier than the evidence allows.

buildtotal ≥ 56

Threshold

All axes ≥ 7 AND total ≥ 56

Every axis has strong evidence. The idea is validated across problem, market, solutions, wedge, distribution, economics, and founder fit. Ready to hand off to Stackforge for stack composition.

Example

Founder with 8 years in the domain, 5 user quotes with URLs, 5 named competitors with dissatisfaction evidence, one clear channel with CAC data.

iterate42–55

Threshold

Total 42–55, no axis ≤ 4

Strong foundation, but two or three axes need more evidence. The core insight is valid — collect the missing signals, re-score. Do not build yet.

Example

Good problem evidence, identified market, but economics is estimated (no real CAC benchmark) and distribution is vague. Score 6, 7, 7, 7, 5, 5, 8 = 45.

scope_cut3+ axes ≤ 5

Threshold

Three or more axes score ≤ 5

The founder is trying to do too much. Pick the single strongest dimension and make it the entire company for the next 12 months. The rest of the axes are speculative.

Example

Problem is crystal clear (9/10), founder fit is strong (8/10), but market definition is fuzzy, solutions are poorly researched, economics is guesswork, wedge is hand-wavy.

pivotany axis ≤ 3

Threshold

Any single axis ≤ 3

One dimension is so weak that fixing it would require redefining the company. The work is not to iterate — it's to go talk to 20 people and find the real pain.

Example

Founder can't name five real competitors (scores solutions 2/10). Either the market is not real, or they haven't looked. Neither is good.

killpivot + core weak

Threshold

Pivot verdict + problem or founder ≤ 3

The foundational signal is missing — either the problem isn't real or the founder has no right to win it. The evidence ledger says stop. Six months saved.

Example

Three user quotes fail the HEAD check (dead URLs), counterfactual quote missing. Problem score: 0. No evidence the pain exists at all.

Exact threshold logic — from the protocol spec

computeVerdict(axes, total):
  if (total >= 56 && no axis score below 6)   → "build"
  if (any single axis score <= 3)             → "pivot"
  if (count(axes where score <= 5) >= 3)      → "scope_cut"
  if (total >= 42 && total <= 55)             → "iterate"
  default                                     → "iterate"

// "kill" is a pivot verdict where problem OR founder scores <= 3
// The roadmap generator uses the verdict to choose milestone templates

Install in 30 seconds

One JSON block in your IDE config.

MarketPrior is published as scale4-mcp-server on npm. Free-tier assessments need no API key — the env block is only required for Pro features.

CUCursor

.cursor/mcp.json

{
  "mcpServers": {
    "scale4": {
      "command": "npx",
      "args": ["-y", "scale4-mcp-server"],
      "env": {
        "SCALE4_API_KEY": "sk4_your_key_here"
      }
    }
  }
}

Free tier works without an API key — remove the env block entirely.

CLClaude Desktop

~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "scale4": {
      "command": "npx",
      "args": ["-y", "scale4-mcp-server"],
      "env": {
        "SCALE4_API_KEY": "sk4_your_key_here"
      }
    }
  }
}

On macOS. On Windows: %APPDATA%\Claude\claude_desktop_config.json.

CLClaude Code

~/.claude.json

{
  "mcpServers": {
    "scale4": {
      "command": "npx",
      "args": ["-y", "scale4-mcp-server"]
    }
  }
}

Edit the user-level config, or pass --mcp-config for a project-scoped config file.

CLCline

VS Code settings.json

{
  "cline.mcpServers": {
    "scale4": {
      "command": "npx",
      "args": ["-y", "scale4-mcp-server"],
      "env": {
        "SCALE4_API_KEY": "sk4_your_key_here"
      }
    }
  }
}

Via the Cline sidebar → MCP Servers → Add server, or add directly to settings.json.

npm: scale4-mcp-server →Full install guide in Docs →

Cross-portfolio handoff

MarketPrior says BUILD →
Stackforge composes the stack.

When a MarketPrior assessment returns a build verdict, the agent can immediately call scale4.stackforge_handoff — which translates the scored assessment into a Stackforge compose_stack intent string. Two MCPs, one agent loop. No copy-paste.

The handoff carries the vertical profile (healthcare, fintech, workforce) from the assessment axes — so a founder who just passed the healthcare-9-axis protocol gets a HIPAA-ready scaffold pre-selected, not a generic starter.

stackforge.dev →Handoff walkthrough →

scale4.score()

70-point scorecard + BUILD verdict

build

↓

scale4.stackforge_handoff()

intent: "multi-tenant SaaS, healthcare" profile: "healthcare"

↓

stackforge.compose_stack()

wrangler.toml + supabase migrations + Clerk JWT + BAA controls

Both MCPs share

HMAC token postureEd25519 signingLicensing serviceCompliance tier cross-sell

Pricing

Per-org. Never per-developer.

Free tier gives you 3 reports per month from the proprietary corpus, no API key required. Pro adds citation verification, PDF export, and full channel-map coverage. All tiers source from the same EDGAR + Crunchbase open + Companies House + press corpus.

Free

$0forever

3 competitive intelligence reports per month, sourced from the MarketPrior corpus. No API key needed.

3 reports / month
Proprietary corpus (EDGAR + Crunchbase open + Companies House + press)
Markdown export
Stateless intake sessions
Citation verification (deep_research_mode)
PDF export + Pitchforge handoff

Generate a free report

Most teams pick

Pro

$29per org / month

Up to 25 generated reports per month with proprietary corpus depth. Citation verification, PDF export, full channel-map coverage.

Up to 25 reports / month with proprietary corpus
Everything in Free
Citation verification (verifyCitations layer)
PDF export
Pitchforge + Stackforge cross-sell handoffs
Shared team workspace + monitored competitors

Upgrade to Pro

Team

$99per org / month

Shared report workspace, monitored competitors, CSV/Notion export. For partnerships, accelerators, and founder collectives.

Everything in Pro
Shared report workspace (team view)
Monitored competitors (daily corpus refresh + alert)
CSV + Notion export
Lite MarketPrior Capital view (3 VC briefs / mo)
Custom verticals + SLA

Upgrade to Team

Enterprise

$890per org / month

Custom vertical corpora (your sector, deep), SLA on the report generator, signed citation-evidence pack for investors and auditors.

Everything in Team
Custom vertical corpora (your sector, deep)
SLA on report generation + corpus freshness
Signed citation-evidence pack
30-day deprecation window on corpus-schema changes
Direct line for corpus + methodology questions

Talk to us

Per-org pricing. Add as many team members as you need at no extra cost. Terms apply.

The competitive landscapebefore you wire the check.

"Great idea — here's the roadmap."

Generic SWOT, zero citations.

Five competitors, three of which don't exist.

Problem hypothesis

Beachhead market

Existing solutions

Wedge

Distribution

Unit economics

Founder fit

Agent calls scale4.start

Agent collects signals, axis by axis

score() unlocks when all axes are complete

Roadmap + handoff emitted

One JSON block in your IDE config.

MarketPrior says BUILD →Stackforge composes the stack.

The competitive landscape
before you wire the check.

MarketPrior says BUILD →
Stackforge composes the stack.