Agent Distribution Engineering
What is Agent Distribution Engineering?
Agent Distribution Engineering is the work of making an existing SaaS product listed, working and chosen inside AI agents: tool design, auth, evals, directory submission and measurement after launch. We don't build agents. We make your product usable by the agents your customers already use. It is the work Invokary does.
Listed, working and chosen
The definition names outcomes, not deliverables. Each one has a measure, so you can tell whether it happened.
- Listed
- A live public listing on every surface you contract, published from your organization. Measured by listing URLs and tier (Community or Verified).
- Working
- Every tool succeeds on valid input and auth works first time. Measured by eval pass rate of 90% or more, tool error rate under 2%, uptime of 99.5% or more.
- Chosen
- The agent picks the right tool, and people find your listing. Measured by tool-selection accuracy of 85% or more, plus listing views and search queries where the platform reports them.
The work it covers
- Tool design
- Choosing the tools an agent sees and writing their names, descriptions, input schemas and hints, so the agent picks the right one. We design 8 to 20 tools around the jobs your users do, not one per endpoint.
- Auth
- Letting each user sign in with their own account, through OAuth that meets the MCP specification and each directory's rules.
- Evals
- Tests that give an agent realistic prompts against your tools and check it picks the right one and gets a correct result, before every submission and after it.
- Directory submission
- Meeting each directory's published requirements and submitting from your own accounts, so the listing is yours.
- Measurement after launch
- Following listing views, installs, tool calls and errors where each platform reports them, and fixing what the numbers show.
What it isn't
- Not generic MCP development: a server is one deliverable; we're accountable for the listing, the review and whether agents pick your tools.
- Not a replacement for AI coding tools: they write code; we own tool design, review rules, evals and submission.
- Not an OpenAPI-to-MCP generator: generators mirror endpoints; we design 8 to 20 tools around real jobs.