Monid is a routed catalog of 2,000+ agent tools across 72 providers

A team wiring an AI agent into real-world data writes an integration per vendor, a billing model per vendor, and an error path per vendor, so the work scales with the number of APIs the agent reaches. Monid, now at 5,804 stars with 16th place on Trendshift's daily ranking, is a one-base-URL layer that an agent can call to reach 2,000+ tools across 72+ providers under a single contract.

The project is at monid-ai/monid, created 26 August 2026 and pushed on 22 October 2026, under an MIT licence. Its own tagline on the README is "OpenRouter, but for agent tools."

What the catalog covers

The categories listed in the README are web search and scraping, people and company enrichment, social platforms, reviews and market data, and video, image and voice generation. A single OpenAI block adds image generation and editing endpoints for GPT Image 2, GPT Image 2.5 Sunburst and GPT Image 2.5 Flare with fixed models and per-token billing.

Three verbs make up the agent-facing API, and the first two are free. discover ranks the whole catalog by the job and returns candidates with live price, health, and observed p50 and p95 latency. inspect shows a single endpoint's schema. run executes a call and is billed per use.

Billing settles on the raw wire response

Every connector declares its own usage model in the definition file: a flat charge per call, a charge per result returned, or a rate per unit such as a thousand characters or a second of video. The engine applies that model to the raw vendor response envelope, before any output mapping, so the amount billed matches what came back from the vendor.

A non-2xx vendor response completes as data in Monid, with the run settling at zero usage. The README names this as a design choice: a vendor error, an unmatched company, an unresolved person each finish with a zero-usage receipt, and the agent reads the outcome from the result envelope. No exception is thrown for the agent to catch.

Writing a connector is a declarative file

A connector describes one provider and its endpoints as a document with a schema. Functions inside the definition are replaced by content-hash references at compile time, and each distinct source is interned once, so the hash doubles as a tamper check. One compiled artifact runs three ways: locally with the author's own vendor key, in CI replayed against fixtures with no network, and in the hosted platform where credentials are injected inside the transport and never enter the engine process.

The declarative shape is on purpose, the README says, because it is work a coding agent can do: pointed at the brief in AGENT.md, the API docs and the Exa worked example, a model can produce the provider, the endpoint schemas and the tests, and CI tells it whether the output compiles against the contract.

The pattern worth copying past a dozen integrations

The pattern Monid names is the one to copy for any team past the first few vendor integrations: a declarative connector, a billing model attached to the connector, replay fixtures that answer whether CI passes without touching the network, and non-2xx treated as data. The quality still depends on the specific connector in the repo, and a bug in it belongs to the user of the agent, not to Monid, so the usual due diligence for a dependency applies.

Source

monid-ai/monid on GitHub and monid.ai.