Rubric v1 — how the score works Snapshot 2026-09-18

The trusted source for your agent

Verified, scored, and structured for autonomous agents. Six published dimensions, one rubric, and no score for sale.

One field, one resource, no account needed.

People are about to stop installing apps. An AI dashboard generates the software someone needs on demand — a calorie tracker today, a travel planner tomorrow — and that software wires up flight data, hotels, maps, payments, and spends its own monthly allowance to finish the job.

When an agent spends money and acts on your behalf, API trust is infrastructure. This registry exists so an agent — or the human choosing for it — can answer one question before the first call: is this API safe, reachable, agent-friendly, and worth the call?

Directory listings that rank by brand fame or ad spend cannot answer that. Every entry here is tried, scored on six published dimensions, and re-checked on a schedule. A household name can score below a smaller rival, and we publish it.

Ten entries, scored

The launch ladder starts with ten APIs and MCP servers that agents already reach for. Each carries one overall score: the weighted sum of the six dimensions below, published to one decimal.

0507085100Anthropic88.5GitHub87.3Stripe84.1OpenAI82.7Auth081.8Google Maps Platform74.1SendGrid73.8Twilio73.1Amadeus60DocuSign49.7Overall score, rubric v1 (0–100)
Ten-entry launch ladder, snapshot 2026-09-18. Scores are illustrative seed data under rubric v1, not measurements of these vendors.
  • 85–100 · Strong agent-ready default
  • 70–84 · Solid with caveats
  • 50–69 · Usable but frictional
  • Below 50 · Avoid as default

What separates them

Read the rows, not just the totals. Ping speed is the easiest dimension to win and the least useful on its own: an endpoint that answers in 90ms is worth nothing to an agent that cannot get a credential. Ease of use and MCP quality are where agent-first products pull ahead.

Anthropic AI & LLM
GitHub Dev tools
Stripe Payments
DocuSign E-sign
Overall score weighted, rubric v1
88.5
87.3
84.1
49.7
Ping speed weight 10
88
95
90
88
Ease of use weight 25
90
92
88
42
API complexity weight 10
80
85
70
35
Quality weight 15
92
95
90
70
MCP quality weight 20
92
70
75
10
Security posture weight 20
85
90
88
72

Accent column: the highest overall in this comparison. All ten entries are in the registry; the four here were chosen to show the spread.

What the score is made of

The weights are published because they are the argument. Ease of use and security outrank raw latency, because that is where agents actually fail. MCP quality carries the same weight as security — a tool surface an agent can call beats a status page it cannot.

0510152025Ping speed10Ease of use25API complexity10Quality15MCP quality20Security posture20Weight toward the overall score, percent
Rubric v1 weights, docs/scoring-rubric.md. A weight change is a version bump and a public changelog entry.

What a high score looks like

Anthropic leads the ladder at 88.5. Against the ten-entry average it is ahead on every dimension, and furthest ahead on MCP quality — the dimension most entries have not built yet.

0255075100Ping speed8887.3Ease of use9077.7API complexity8066Quality9284.5MCP quality9258.7Security posture8581.6Dimension score, 0–100
Dimension scores for the leader and the arithmetic mean of the ten entries.
  • Anthropic · 88.5
  • Ten-entry average

Built for agents

The same truth is served three ways: the page you are reading, a JSON API, and an MCP server. An agent should be able to check a score mid-task instead of scraping HTML.

$ curl https://trustedapis.com/api/v1/apis?limit=10
$ curl https://trustedapis.com/mcp \
    -d '{"method":"verify_api","params":{"slug":"stripe"}}'

MCP methods: verify_api, get_api_score, list_category, report_incident. Agents that read files first can start at /llms.txt.

Get verified

Listings are free to apply for. Vetting is manual: a real trial account, the smallest path to a working credential, and at least three representative calls. Vendors can contest methodology and supply evidence — corrections are published, not quietly applied.

Ten entries, rubric v1, snapshot 2026-09-18. Scores are illustrative seed data, not measurements of these vendors. The rubric is at /verification; the rankings are never for sale.