State of Agent-Readiness: March 2026

We scanned 190+ of the most popular SaaS tools and APIs. We graded each one on 8 agent-readiness criteria. Here's what we found.

The headline: most tools aren't ready

The average AgentGrade across all tools is C+ (5.95/10). Only 13% scored B+ or above. Just 3 tools earned an A grade. Most tools were built for humans clicking buttons, not agents making API calls.

What separates the top-scoring tools

The highest-scoring tools share common traits:

• OpenAPI specs that are accurate and complete
• API keys with fine-grained scoped permissions
• Sandbox/test modes for safe experimentation
• Rate limits clearly communicated via headers
• Compact response payloads (field selection, no bloat)

OpenAI (A, 8.26) leads the pack with comprehensive REST API, API key auth, clear documentation, and SDKs in every major language. GitHub (A, 8.02) and Anthropic (A, 8.02) follow with excellent API design, scoped tokens, and broad access methods including MCP servers.

The biggest friction points

1. OAuth-only auth (affects 40% of tools): Many tools require OAuth flows that need a browser. Agents can't click through consent screens.
2. No field selection (affects 70%): Most APIs return everything. No way to say "just give me the name and status."
3. No sandbox mode (affects 60%): Agents need to experiment safely. Without test mode, every action is live and irreversible.
4. Missing OpenAPI specs (affects 55%): Without a machine-readable API spec, agents rely on scraping documentation pages.

Category leaders

What needs to change

For SaaS companies serious about the agent economy, three changes have the highest impact:

1. Add API key auth alongside OAuth (low effort, high impact)
2. Add field selection to reduce response sizes
3. Publish an OpenAPI spec (even auto-generated is better than nothing)

We'll publish updated numbers quarterly. Browse all scores at agenttool.sh.