iAgentMart Agent API

iAgentMart is a focused directory for resources an AI agent can install, invoke, or control through a machine-readable interface. It intentionally contains only four resource types.

Canonical collection
https://iagentmart.com/api/resources

Four resource types

TypeMeaning
mcpModel Context Protocol servers and connectors exposing tools, prompts, or resources.
skillReusable agent instructions, workflows, or capability packs.
pluginInstallable extensions and integrations for agent runtimes.
agent_appApplications intentionally optimized for AI agent control.

Every resource also has exactly one required, type-specific subcategory. Each type includes other for resources outside the maintained list. The collection response returns the complete controlled taxonomy and labels.

Agent App eligibility

An application is eligible only when its repository documents a concrete Agent interface or Agent optimization. Examples include an Agent API, semantic element identifiers, observable structured state, stable actions, MCP tools, or CLI control designed for automation.

Human-only software is out of scope
An app is not eligible merely because an AI model was used to build it or because it contains an ordinary chat feature.

Machine discovery

GET/.well-known/agentmart.json

Returns canonical endpoints, accepted resource types, filters, schema identifiers, and recommended agent behavior. A concise text version is also available at /llms.txt.

Resource detail

GET/api/resources/{resource_id}

Open the detail before installation or invocation. The response includes the public GitHub repository, agent optimization evidence, interfaces, capabilities, platforms, license, curator, and update time.

Response schema

{
  "schemaVersion": "agentmart.resource.v2",
  "resourceType": "agent_app",
  "subcategory": "browser_web",
  "subcategoryLabel": "Browser & Web",
  "repository": { "owner": "example", "name": "agent-browser", "url": "https://github.com/example/agent-browser" },
  "agentOptimized": true,
  "agentEvidence": "Documented Agent API with semantic page state and structured actions.",
  "interfaces": ["agent_api", "semantic_state"],
  "capabilities": ["browser_control"],
  "platforms": ["macos", "windows", "linux"]
}

Collection responses additionally include normalized query data, taxonomy, agent hints, and cursor pagination.

Repository submission

POST/api/resources

Submission requires a GitHub browser session. A curator may submit any public, non-archived GitHub repository; ownership is not required. Required fields are repositoryUrl, resourceType, subcategory, and agentEvidence.

{
  "repositoryUrl": "https://github.com/example/agent-browser",
  "resourceType": "agent_app",
  "subcategory": "browser_web",
  "agentEvidence": "The repository documents an Agent API with stable action IDs.",
  "interfaces": ["agent_api", "structured_actions"],
  "capabilities": ["browser_control"],
  "platforms": ["cross_platform"]
}

Duplicate identity

Repository identity is case-insensitive and canonicalized to github.com/owner/repository. Repository, release, archive, tree, and file URL variations resolve to the same identity. Submitting an existing repository returns the existing resource with duplicate: true and does not create or alter another row.