Make your product usable by AI agents: the one-person distribution trick

You have a product and no marketing team. The durable channel is a surface an AI agent can call, schedule, or publish through. It is a margin play: once the surface works, it can be called without a human click.

Postiz, the social scheduler, had about $332,000 in Stripe-verified MRR on October 29, 2026, with more than $1.6 million in all-time revenue. Nevo David built it as a one-person, investor-free open-source scheduler for agents. It gave external tools three ways to work with it: a public API, an n8n node, and a CLI that agents can use to plan, create, and send posts across more than 20 networks. Its open-source edition could run on a self-hosted server at no cost under the AGPL-3.0 license. For an open-source SaaS, that free path matters because it lowers the cost of trying the product.

Postiz's Stripe-verified MRR stood at $60,731 on March 2, 2026 and $87,530 on April 2, 2026. By October 12, 2026, it was $262,899, with 7,683 active subscriptions. A large share of recurring revenue is processed through Whop, so it is not reflected in the Stripe MRR totals.

Agents call tools, not pages

In 2026, AI agents are already operating in production for real teams, handling meetings, support-ticket triage, lead qualification, and code delivery, while 76% of companies are using automation to make routine operations more consistent. An agent does not browse your homepage. It asks for a tool it can call. If your product cannot be called, it is invisible to that buyer.

MCP, introduced by Anthropic in late 2024, has become the default protocol for agents to call external tools. n8n's 2026 MCP support lets external AI platforms, including Claude and Lovable, call its workflows. AI buyer agents compare suppliers on price, stock, and terms, so scattered sales systems can lose deals. A solo operator cannot outspend a sales team. You can out-prepare one by giving agents one clean answer.

An agent buyer needs a contract: the inputs it accepts, the outputs it returns, the cost, and the failure behavior. The contract is the docs, the auth, and the limits. Put those three in one place.

Three surfaces make the product callable

Keep the checklist: public API, workflow node, and CLI, each with docs, auth, and rate limits.

• Public API: expose the core action your product performs. Write docs for endpoints, request shapes, response shapes, and error codes. Add API-key authentication, and set rate limits per key and per minute. Do not expose raw database tables or undocumented endpoints; agents will find the gaps, and a broken call is worse than no call. Done looks like an agent creates a resource, receives a stable identifier, and can retry without a human clicking through the interface.
• Workflow node: add a node for a platform such as n8n. Document the node's input fields, output fields, and credential setup. Store authentication in the workflow platform, not in the workflow itself. Respect upstream rate limits, and surface throttling errors in a way a non-engineer can read. Finished means a user drags the node into a workflow, connects it to a trigger, and runs it without writing code.
• CLI: ship a command-line interface for the same core action. List commands, flags, examples, and exit codes. Use environment variables or a config file for secrets, and keep them out of command history. Add client-side throttling, and return clear errors when the service is busy. You know it works when an agent runs the command from a script, parses the output, and continues the task.

Trust is what keeps agents coming back

Agents do not forgive flaky tools. They retry, abandon, or choose another supplier. Make the docs current, the auth predictable, and the rate limits visible. If a surface changes, version it. If a limit is hit, say so plainly.