Methodology · Rubric v2
How we score.
Magpie AI is the opinionated, stage-aware AI tool directory for early-stage startup founders building their company's foundational stack: scored against a real rubric, agent-queryable via MCP, and free of paid placement.
Most AI tool directories rank tools by superficial signals: vendor marketing copy, paid placement, or raw G2 ratings. None of them are honest about whether a tool is actually a good fit for a founder with a real budget and a real job to do.
Magpie AI scores every tool against six dimensions. The result is a 1.0-5.0 overall score and a recommendation badge that tells you, plainly, whether this tool is worth your time and money.
The six dimensions
SME fit
25%Pricing accessibility, total cost of ownership at scale, buyer-journey friction.
We dock tools that hide pricing, force enterprise sales, or stack a 10× cliff between tiers. We reward genuine free tiers and predictable per-seat pricing under $30/seat.
Job-to-be-Done Clarity
15%How specifically does this tool solve a named, measurable job?
Tools that claim to “do everything with AI” rarely do anything well. We reward focused tools that name a job and a measurable outcome.
Integration & Agent-Readiness
25%Public API, MCP server, native integrations with the everyday SaaS stack.
Tied with SME fit for the highest weight. A tool that can’t be used by an agent today is a tool that won’t be used tomorrow, by anyone, solo operator or scaling startup.
Trust, Stability & Lock-in
15%Company age, data portability, contract terms.
Founded in 2024 with no data export and a 3-year contract minimum? We mark it down. Three-year-old, profitable, with one-click CSV export? We mark it up.
Momentum & Quality Signals
15%Rating velocity, community engagement, docs quality, changelog cadence.
A 4.5 climbing toward 4.7 is a healthier signal than a 4.7 falling to 4.5. We try to surface direction, not just absolute numbers.
Compliance & Data Practices
5%SOC 2, GDPR, data residency, DPA availability.
Default weight is low because most solo operators and small teams don’t need SOC 2 Type II. For finance / legal / HR / health categories, this dimension scales up to 20-25%.
Default weights apply to most categories. Finance / legal / HR / health categories scale up Compliance to 20-25% at the expense of JTBD.
Auto-disqualifiers
Any one of these strips the tool's recommendation badge entirely. The low overall score becomes the only user-facing signal. Auto-DQs are absolutes, not weighted signals.
- Buyer journey is demo-only (no way to evaluate before a sales call)
- No data export option
- No public roadmap or changelog in 12+ months
- Dead or sunset (status=sunset, or no updates in 12+ months)
- Sketchy data practices (no privacy policy, or vendor claims customer data ownership)
- Not strictly AI-native (legacy tool with an AI label slapped on)
Recommendation badges
The minimum-dimension floor matters: a tool scoring 5/5/5/5/5/1 averages 4.3, but the 1 in any dimension is a red flag worth surfacing. The badge logic catches that and flags it as Consider with caveats rather than letting the high average hide the gap.
Who Magpie is for
Built for founders.
Magpie is opinionated about who it serves. The rubric, the stage filters, and the editorial picks are tuned for the people below.
- Pre-seed and seed-stage founders making foundational tooling decisions. The primary audience.
- Solo operators and indie hackers picking tools they'll live with.
- Series A and growth-stage operators who started with Magpie at seed and want the same opinionated scoring as their stack expands.
- Anyone shipping AI-native products who needs to know what plugs into Claude, Cursor, or their own agents.
Who scores these tools
Jack Phillips
Founder & editor, Magpie AI
I run Magpie. Every score on this site, every comparison, every category intro is something I've walked through myself. Magpie exists because the existing AI tool directories are SEO farms with paid placement; the founders I talk to deserve a directory with an actual point of view.
If a score looks wrong, the rubric is public and the math is in docs/rubric.md. Tell me where the math is off and I'll fix it.
Read the full spec
The complete rubric (formulas, worked examples, dimension inputs by sheet column) lives in docs/rubric.md on GitHub. The validation agent reads the same document; it's the single source of truth.
Found a tool that should be in the directory? Submit it.