The measure of
agent-readiness
Developer tools have a new class of customer. Already today, AI coding agents are choosing and using tools at a higher rate than humans.
Is your tool ready for them?
The Standard at a glance
The Zaira Standard defines what "Agent Ready" means: 71 criteria across all domains of a tool's function, based on broad industry research, delivered in an open methodology.
What's measured
Base Standard applies to all tools. Additional modules activate based on tool complexity.
How it's scored
Scoring gradients are built to reflect realistic use cases and how agents work today.
Where you stand
The Zaira Score is the weighted percentage across all applicable modules.
The best tools are independently evaluated. See what it takes to get Zaira Certified.
One standard, right-sized for every tool
"Developer tools" is a broad category where different tools have different needs. Agent-readiness looks very different for a simple utility library vs. a payment processing API. The Standard is intentionally modular: a base evaluation applies to every tool, with additional complexity modules activating based on what your tool actually does. You're only measured on what's relevant.
Base Standard
Applies to every tool, always. The foundation everything else builds on.
Documentation quality, machine-readable formats, error handling, versioning, security posture, maintenance signals, licensing.
| Module | # | When | Coverage | |
|---|---|---|---|---|
PI | Programmatic Interface | 16 | Tool exposes an API, SDK, or MCP server | SDK quality, API design, type safety, error contracts |
NS | Network Service | 8 | Tool is a hosted or remote service | Uptime, error responses, rate limiting, latency |
WO | Write Operations | 4 | Tool creates, modifies, or deletes data | Idempotency, rollback, conflict handling |
AU | Authentication | 4 | Tool requires credentials or tokens | Key management, token flows, non-interactive auth |
CLI | CLI | 4 | Tool has a command-line interface | Exit codes, machine-readable output, scriptability |
| Module | # | When | Coverage | |
|---|---|---|---|---|
PM | Payments | 4 | Tool processes payments | Test mode parity, webhook verification, PCI scope |
CM | Communications | 3 | Tool sends messages or notifications | Delivery status, templates, channel abstraction |
DB | Databases | 4 | Tool stores or queries data | Migrations, connection pooling, schema introspection |
HI | Hosting | 3 | Tool deploys or hosts applications | Deploy automation, environment parity, scaling |
AP | Auth Providers | 3 | Tool manages user identity | Non-interactive flows, MFA, session management |
FL | Frameworks | 3 | Tool scaffolds or structures applications | Scaffolding, conventions, plugin architecture |
How modules activate
Your tool's characteristics determine which modules apply.
Complex tool: payment API
Has an SDK, runs as a network service, handles write operations, requires authentication, operates in the payments domain.
Base (15) + PI (16) + NS (8) + WO (4) + AU (4) + PM (4) = 51 criteria
Simple tool: open-source library
No API, no network calls, no auth. A library you install and call. Only the Base Standard applies.
Base (15) = 15 criteria
Want to know what this looks like
in practice?
The Best Practices Guide walks through every criterion with concrete and realistic implementation patterns with real examples. The same guidance ships as an installable Agent Skill your coding agent can apply to your repo.
Built on evidence, not opinion
The Zaira Standard wasn't designed in a vacuum. Every criterion traces back to published research, real-world incident data, or empirical measurement of how agents actually interact with developer tools. The findings shaped what we measure and why.
Examples from the research
Each finding below directly shaped a criterion in the Standard.
43% of analyzed MCP servers have command injection vulnerabilities. In one study, every one of 1,404 real-world email agents evaluated was eventually hijacked via prompt injection.
Input sanitization is a MUST gate: a tool that scores perfectly on documentation but fails it cannot reach Agent Ready or Agent Native. Security is non-negotiable.
Agent error recovery nearly triples when tool failures are systematically curable. 97.1% of MCP tool descriptions have quality defects. Tool-use examples lifted accuracy from 72% to 90% in Anthropic's internal testing.
Error handling and tool description quality are weighted as Critical (2x) in scoring. These aren't nice-to-haves. The data shows they're the difference between an agent that recovers and one that fails.
Agents given concise, curated context files used roughly 40% fewer tool calls and tokens per task. But more documentation is not better. Accuracy degrades 30%+ when relevant information sits in the middle of large context.
Documentation criteria evaluate structure and extractability, not volume. The Standard rewards answer-first layouts, semantic sections, and machine-readable formats over comprehensive reference dumps.
Additional metrics and research findings can be found throughout the Standard and Best Practices Guide.
Open and stable
The Zaira Standard is public and stable. Revisions are made only when durable shifts in agent-tool interaction warrant them, not in response to short-term trends. Certified tools receive at least 30 days advance notice before any version change takes effect, giving vendors time to adapt ahead of the transition.
The first version is purposefully v0.9 to signify the expected tuning based on pilot evaluation feedback before v1.0 is published.