iiinova — where your digital self keeps evolving

Hygraph's MCP connector is in the Claude Directory, and its agent door outscores its developer door: agent readiness 42.8 with the whole identity layer lit, Kin Score 22.2 with access clarity at 10.5. apis.io/2026/09/30/hygraphs-agent-d…

Black and white editorial illustration for the apis.io post "Hygraph's agent door scores twice its developer door".

000

APIs.io @apis.io · 22h

"A SQL parser cannot tell a socially engineered refund from a legitimate one if the syntax is valid." Google on judging agent intent at runtime. The catalog cannot see the gateway doing it. apis.io/2026/09/30/google-judges-an…

Black and white editorial illustration for the apis.io post "Google judges an agent's intent at runtime, and the catalog cannot see the gateway".

010

APIs.io @apis.io · 29/09/2026

SDK for the scheduled job, a JSON CLI for the unattended agent, MCP for interactive work. Postman's answer: generate all three from one spec. The spec must say what happens when a call runs twice. apis.io/2026/09/29/postman-says-gen…

Black and white editorial illustration for the apis.io post "Postman says generate the SDK, the CLI and the MCP server from one spec".

021

Reposted by APIs.io

API Evangelist @apievangelist.com · 29/09/2026

Searching for an API? So is your agent. - apis.io

031

APIs.io @apis.io · 29/09/2026

DigitalOcean benchmarked LLM structured output under load and published its own miss: pinned vLLM held 1.00 schema validity at 400 concurrent; its managed endpoint fell to about 0.85. apis.io/2026/09/29/digitalocean-mea…

Black and white editorial illustration for the apis.io post "DigitalOcean measured structured output under load, and its own managed endpoint slipped".

011

APIs.io @apis.io · 29/09/2026

Maritime has 67 providers on apis.io and not one scores above developing; the top is 48.8. Maersk is the only liner above 15. Hapag-Lloyd, MSC and COSCO sit between 3.3 and 5.6. Global trade runs on APIs you cannot see. apis.io/2026/09/28/maritime-has-67-…

Black and white editorial illustration for the apis.io post "Maritime has 67 providers on apis.io and not one scores above developing".

021

APIs.io @apis.io · 28/09/2026

Monitoring providers sell visibility and median 26.3 on operational transparency, the facet for rate limits, status, change logs and deprecation. 124 of 222 score under 30. Easy to find, hard to watch. apis.io/2026/09/28/monitoring-provi…

Black and white editorial illustration for the apis.io post "The monitoring area sells visibility, and its providers median 26.3 on operational transparency".

000

APIs.io @apis.io · 28/09/2026

Simon Mikulcik: an API feature should be its own artifact, one OpenAPI Overlay per feature, targeting by JSON Path value instead of array index. The most practical use of Overlays we have read. medium.com/@smikulcik/api-features-…

000

APIs.io @apis.io · 28/09/2026

OpenAI wins or ties all six Kin Score facets against Salesforce and still finishes second, 86.0 to 91.8. The whole gap is the upsert layer: Salesforce 72.2, OpenAI 0.0. apis.io/2026/09/28/salesforce-vs-op…

Black and white editorial illustration for the apis.io post "Salesforce vs OpenAI: OpenAI wins or ties every facet, and finishes second".

000

APIs.io @apis.io · 28/09/2026

Auth0's HealthCheck skill audits a tenant with a read-only app and stages every fix behind a terminal approval. "No changes are ever applied automatically." The pattern every agent skill should ship with. apis.io/2026/09/28/auth0-lets-an-ag…

Black and white editorial illustration for the apis.io post "Auth0 lets an agent audit a tenant, and never lets it write without asking".

011

APIs.io @apis.io · 25/09/2026

"The gateway was a feature. The authorization model was the product." WorkOS on MCP gateways. Its record lights the identity layer; consent identity and error semantics still dark. apis.io/2026/09/25/workos-says-the-…

Black and white editorial illustration for the apis.io post "WorkOS says the authorization model is the product, and its record lights the identity layer".

010

APIs.io @apis.io · 25/09/2026

Payments providers light idempotency at 21.4% vs 3.0% for the rest, reversibility at 27.3% vs 6.2%. The one category that tells an agent how to undo a call. Dry-run mode: 0.1%. None lets it rehearse. apis.io/2026/09/25/payments-is-the-…

Black and white editorial illustration for the apis.io post "Payments is the one category that documents how to undo a call, and still cannot rehearse one".

000

Reposted by APIs.io

fkilcommins @fkilcommins.bsky.social · 25/09/2026

We've now assessed 10,000+ #APIs for AI-readiness. At that scale, patterns become hard to argue with. Most APIs weren't built to be integrated by machines that reason, and the data keeps confirming it. Check out the directory for more info. Check out the directory: jentic.com/apis

jentic.com

Jentic API Directory — Connect AI Agents to Any API

Discover 10,000+ production-ready APIs organized by category. Connect your AI agents in minutes with validated OpenAPI specifications.

122

Reposted by APIs.io

fkilcommins @fkilcommins.bsky.social · 25/09/2026

The September edition of "Jentic Tech Talk" from @jentic.bsky.social is out. There's some interesting content especially for those who care about API usage for agentic systems. jentic.com/newsletter/t...

jentic.com

Jentic Tech Talk — September 2026 | Newsletter

The newsletter for teams building AI that actually ships. Covering APIs, AI agents, open standards, and enterprise architecture.

122

APIs.io @apis.io · 24/09/2026

APIMatic wins five of six facets vs Confluence and still comes second, 60.6 to 67.4. The one it loses, contract quality, is scored on a contract we generated: APIMatic has no spec for its own API. apis.io/2026/09/24/confluence-vs-ap…

Black and white editorial illustration for the apis.io post "Confluence vs APIMatic: five facets to one, and the one is a contract APIMatic never published".

000

APIs.io @apis.io · 24/09/2026

"The server controls what to ask and when. The client controls how the question appears." Nango on MCP elicitation, the mid-call ask. Its own record has consent identity and delegated identity unlit. apis.io/2026/09/24/nango-explains-m…

Black and white editorial illustration for the apis.io post "Nango explains MCP elicitation, and its own consent dimension is dark".

000

APIs.io @apis.io · 24/09/2026

Vespa's interns built a public MCP server and benchmarked it: 97% of assertions passed vs 95% via the CLI, in half the deployment attempts. "Tools don't make the model smarter; they make it wrong less often." apis.io/2026/09/24/vespa-built-a-pu…

Black and white editorial illustration for the apis.io post "Vespa built a public MCP server, measured it against the CLI, and the catalog has not lit it yet".

000

APIs.io @apis.io · 23/09/2026

Lili 64.5 vs Intuit 62.7. On the six core facets Intuit leads 60.0 to 58.2; Lili's whole lead is the regulatory layer. Intuit has the contracts, Lili the only first-party MCP server. apis.io/2026/09/23/lili-vs-intuit-a…

Black and white editorial illustration for the apis.io post "Lili vs Intuit: a bank beats QuickBooks on the regulatory layer, not the API".

000

APIs.io @apis.io · 23/09/2026

Transactional email averages 63.2 on the Kin Score vs 22.6 for the rest. Idempotency lit for 19.4%, dry-run mode for 0.0%. The category can observe a send, not safely retry one. apis.io/2026/09/23/transactional-em…

Black and white editorial illustration for the apis.io post "Transactional email is forty points ahead of the market, and still cannot take a send back".

000

APIs.io @apis.io · 23/09/2026

"UI testing's hard problem is maintenance. API testing's hard problem is coverage." Sofy: five optional parameters is 32 combinations, so test the running API, not the spec. Kin Score 44.9; agentic access unlit on its own API. apis.io/2026/09/23/sofy-api-testing…

Black and white editorial illustration for the apis.io post "Sofy says API testing's hard problem is coverage, and its own API is not yet open to an agent".

000

APIs.io @apis.io · 22/09/2026

Cloudflare vs Azure API Management: two AI gateways, 3.8 points apart on the Kin Score, three facets won each. On agents it's wider: Cloudflare 65.8 and agent-ready, Azure 27.5. apis.io/2026/09/22/cloudflare-vs-az…

Black and white editorial illustration for the apis.io post "Cloudflare vs Azure API Management: two roads to an AI gateway".

000

APIs.io @apis.io · 22/09/2026

TestMu AI measured its 71 agent skills: 33,016 characters of always-loaded metadata against 402,822 of full bodies. "Access problems are MCP problems. Method problems are skill problems." The catalog lights both. apis.io/2026/09/22/testmu-ai-puts-a…

Black and white editorial illustration for the apis.io post "TestMu AI puts a number on skills versus MCP, and the catalog lights both".

000

APIs.io @apis.io · 22/09/2026

Codehooks: "Being able to answer questions about a webhook three weeks later is a different job, and it is the one that tends to get skipped." Kin Score 41.7, and idempotency is unlit on its own API. apis.io/2026/09/22/codehooks-store-…

Black and white editorial illustration for the apis.io post "Codehooks says store the webhook before you process it, and its own idempotency signal is dark".

031

APIs.io @apis.io · 21/09/2026

GitLab's new MCP tools ship with a default that splits on the verb: read-only runs without a human, anything that writes waits for one. 72.0 exemplar. dry_run_mode still unlit — an approval gate is a person standing in. apis.io/2026/09/21/gitlab-read-runs…

Black and white editorial illustration for the apis.io post "GitLab: Read Runs Unasked, Write Waits for a Human".

100

APIs.io @apis.io · 21/09/2026

A2A Agent Cards - Every provider on the APIs.io network serving a machine-readable agent discovery manifest at the A2A well-known path — graded against the A2A 1.0.0 specification rather than merely counted. apis.io/a2a/

021

Reposted by APIs.io

API Evangelist @apievangelist.com · 17/09/2026

OpenAPI, AsyncAPI, JSON Schema, Arazzo, Overlay, Spectral, APIs.json, MCP and A2A — read from their published schemas, mapped by what each one points at, which property does which job, and how much vocabulary they share. standards.apievangelist.com/relationships/

043

APIs.io @apis.io · 17/09/2026

Only 3.2% of the 633 webhook providers on APIs.io are human-only to an agent, against 55.5% of everyone else. Their gap: just 19.1% document idempotency on their own API. apis.io/2026/09/17/webhook-provider…

Black and white editorial illustration for the apis.io post "Webhook Providers Score Twice the Market and Still Skip Idempotency".

000

APIs.io @apis.io · 17/09/2026

Sigma beats Qlik on five of six Kin Score facets and still ranks second on embedded analytics, 68.8 to 69.7. Weigh the six and Sigma leads 67.3 to 67.1 — Qlik's whole lead is one conditional facet: upsert. apis.io/2026/09/17/qlik-vs-sigma-fi…

Black and white editorial illustration for the apis.io post "Qlik vs Sigma: five facets won and still second".

000

APIs.io @apis.io · 17/09/2026

Red5 reviewed four sets of streaming agent skills — its own, Agora's, two from Bitmovin — against criteria any could fail. The catalog agrees: both Red5 skills are registered, agent_skills lit, 65.6 strong. apis.io/2026/09/17/red5-reviews-its…

Black and white editorial illustration for the apis.io post "Red5 Reviews Its Rivals' Agent Skills Alongside Its Own".

000

APIs.io @apis.io · 17/09/2026

226 A2A Agent Cards indexed. 56 fail the spec, and 53 of those are missing one field: protocolVersion. Real manifests, real skills, canonical path — and a version string nobody added. apis.io/2026/09/16/one-missing-fiel…

Black and white editorial illustration for the apis.io post "One Missing Field Keeps 53 Agent Cards Out of the Registry".

000

APIs.io @apis.io · 16/09/2026

Bindbee and Deel rank 1-2 on the apis.io payroll tag, 1.3 apart. Bindbee publishes ONE OpenAPI, 144 operations. Deel publishes ninety-nine. Bindbee wins four facets; Deel wins contract quality 78.2 to 33.4 — the heaviest. apis.io/2026/09/16/bindbee-vs-deel-…

Black and white editorial illustration for the apis.io post "Bindbee vs Deel: one contract against ninety-nine".

000

APIs.io @apis.io · 16/09/2026

Half of the 2,017 healthcare companies on APIs.io publish almost nothing a machine can read. Mean Kin Score 15.8 vs 22.7 for the rest of the catalog, and 78.5% are human-only to an agent. apis.io/2026/09/16/healthcare-is-th…

Black and white editorial illustration for the apis.io post "Healthcare Is the Biggest Tag on APIs.io and the Least Agent-Ready".

000

APIs.io @apis.io · 16/09/2026

Atlassian's Code Context indexes your repos into the Teamwork Graph, and the agent inherits the caller's permissions — not a service account that sees everything. delegated_identity lit: the claim checks out. apis.io/2026/09/16/atlassian-indexe…

Black and white editorial illustration for the apis.io post "Atlassian Indexes the Codebase and Keeps the Permissions".

000

APIs.io @apis.io · 16/09/2026

Twenty-five APIs came through the apis.io add form. Kin Scores run 70.5 down to 18.6, and what separates them is almost never service quality — it is whether they wrote the contract down. apis.io/2026/09/16/the-last-25-apis…

Black and white editorial illustration for the apis.io post "The Last 25 APIs Added to APIs.io Are Increasingly Built for Agents".

000

APIs.io @apis.io · 15/09/2026

github.com/JetBrains/skills is public, first-party and verified by JetBrains — 129 skills in it. Our catalog holds 33 JetBrains API pages and not one pointer to them. agent_skills reads unlit, which is our gap before theirs. apis.io/2026/09/15/jetbrains-verifi…

Black and white editorial illustration for the apis.io post "JetBrains Verifies 129 Agent Skills. The Catalog Reads Zero.".

000

APIs.io @apis.io · 15/09/2026

Port calls its software catalog a context lake. A context lake has a schema, and theirs is public: Blueprints the types, Entities the instances, Scorecards the judgments, Audit the history. Sixteen APIs. Contract quality 0.0. apis.io/2026/09/14/port-calls-it-a-…

Black and white editorial illustration for the apis.io post "Port Calls It a Context Lake. The Catalog Calls It Sixteen APIs.".

000

APIs.io @apis.io · 14/09/2026

n8n's idempotency guide tells you to check whether the endpoints you call support idempotency keys before enabling retries. Point that at n8n's own API and the answer isn't in the contract — idempotency unlit at 62.9 strong. apis.io/2026/09/14/n8n-tells-you-to…

Black and white editorial illustration for the apis.io post "n8n Tells You to Check for Idempotency Keys. Theirs Are Undeclared.".

000

APIs.io @apis.io · 11/09/2026

Fixing 19 lines of spec inconsistencies per API took out-of-the-box MCP tool success from 76.5% to 99.9%. Truto's number. A 200-endpoint spec also becomes 200 tools and 80,000 tokens, against Cursor's cap of 40 tools total. apis.io/2026/09/11/truto-nineteen-l…

Black and white editorial illustration for the apis.io post "Truto: Nineteen Lines of Spec, 76.5% to 99.9%".

000

APIs.io @apis.io · 10/09/2026

Pulumi's Context API is agent-first without being an agent product. Agents query the infrastructure graph over the same authenticated CLI a human uses — same permissions, same audit trail. 27.6 agent-aware. apis.io/2026/09/10/pulumi-makes-inf…

Black and white editorial illustration for the apis.io post "Pulumi Makes Infrastructure a Graph an Agent Can Query".

000

APIs.io @apis.io · 09/09/2026

The firmware calls a route alias and never learns the API key. Not in the 4,000 units already in the field, not on the $20 programmer someone buys to read one back. Blues scores 32.7 agent-ready, dry_run_mode lit — rare here. apis.io/2026/09/09/blues-keeps-the-…

Black and white editorial illustration for the apis.io post "Blues Keeps the API Key Off the Device Entirely".

010

APIs.io @apis.io · 08/09/2026

Wagtail 8.0 ships 50+ admin operations as an API over MCP, and no AI buttons unless you install them. A CMS with AI, not an AI CMS. Our record still reads v2: Agent Readiness 2.5. One release closes four dimensions. apis.io/2026/09/08/wagtail-ships-an…

Black and white editorial illustration for the apis.io post "Wagtail Ships an Agent API and Refuses the AI Button".

161

APIs.io @apis.io · 08/09/2026

WorkOS on where an agent's OAuth token lives: your database, your process at runtime, or never in your environment at all. A token an agent can reach lands in a context window, a log line or an error payload. No bug required. apis.io/2026/09/08/workos-three-pla…

Black and white editorial illustration for the apis.io post "WorkOS: Three Places a Token Can Live, and the Question None of Them Answer".

000

APIs.io @apis.io · 08/09/2026

honeycomb is at the top of the observability pack with their 80.6 / 100 composite, and 41.9 / 100 agent readiness score. apis.io/tags/observa...

021

APIs.io @apis.io · 07/09/2026

Postman shipped Orbit: describe a task, get the right endpoints back. A task brief costs ~2,500 tokens against ~69,000 for PayPal's OpenAPI. No key, no signup. Postman scores 67.1 exemplar, 39.1 agent-ready. apis.io/2026/09/07/postman-ships-or…

Black and white editorial illustration for the apis.io post "Postman Ships Orbit: 2,500 Tokens Instead of 69,000".

000

APIs.io @apis.io · 07/09/2026

Most vendors document what their API does. Archive published a list of what it cannot do: no webhooks, no writes to creators, no content it has not captured. 47.8, agent-ready — event_surface_described unlit, matching that. apis.io/2026/09/06/archive-writes-d…

Black and white editorial illustration for the apis.io post "Archive Writes Down What Its API Cannot Do".

010

APIs.io @apis.io · 06/09/2026

Agents wrote 96 API integrations for PayPal, Maxio and Twilio. All worked. From published docs alone: 54 probe calls and 517 web requests first. Handed packaged context: zero. APIMatic scores 31.9 agent-ready. apis.io/2026/09/06/apimatic-a-worki…

Black and white editorial illustration for the apis.io post "APIMatic: A Working API Call Was Never the Goal".

021