Connector & Tool Catalog

Version: 1.1.1 | Last verified against source: 2026-08-28

This is the canonical map of how tools enter a OneRingAI agent. It answers three questions that are easy to conflate:

1. Which connectors can OneRingAI configure?
2. Which connectors receive purpose-built tools?
3. Which tools exist without a connector?

For authentication fields and credential URLs, see the complete vendor reference. For tool parameters and longer examples, see Tools & Function Calling.

The four tool sources

agent.tools and agent.context.tools are the same ToolManager instance, so permissions, circuit breakers, execution plugins, enable/disable state, and events apply regardless of where a tool came from.

What every service connector gets

When a connector has a baseURL, ConnectorTools.for(name) creates a generic authenticated API tool:

{connectorName}_api({ method, endpoint, body?, queryParams?, headers? })

The tool:

• uses the connector as the single source of truth for authentication;
• prevents the LLM from overriding protected auth headers;
• supports GET, POST, PUT, PATCH, and DELETE;
• binds userId and optional accountId through ToolContext;
• inherits connector retry, timeout, circuit-breaker, and metrics behavior;
• uses session permission scope by default.

If a specialized factory exists for the connector's serviceType, its tools are appended automatically. All returned tool names are prefixed with the actual connector name, so a base tool such as send_email becomes work_google_send_email for a connector named work-google.

First-party specialized connector packs

Names below are base names. ConnectorTools.for() prefixes every name with the connector name and also includes {connectorName}_api when a baseURL is present.

Web search and scraping

web_fetch is different: it is a connector-free built-in for ordinary HTTP fetching and document conversion. Use a search connector to discover URLs and a scrape connector when sites require a rendering or anti-bot service.

AI media connectors

AI model connectors use vendor, rather than an external-service template. Their media factories are also discovered by ConnectorTools.for():

The external Google Workspace template uses service type google-api; the AI provider connector uses vendor google. They are intentionally separate so a Gemini key and a user's Google Workspace OAuth identity do not get conflated.

All 50 external-service templates

Every template below currently declares a baseURL, so it receives the generic authenticated API tool. “Dedicated” means the specialized pack is added too.

Templates configure authentication and sensible endpoint defaults; they do not claim to model every operation of every vendor. For generic-only services, use the authenticated API tool, call Connector.fetch() directly, or register a domain-specific factory with ConnectorTools.registerService().

Generated connector-free built-ins

ToolRegistry.getBuiltInTools() currently exposes 39 generated entries in eight categories. These are the stable registry entries; optional plugin and connector tools are intentionally discovered at runtime instead.

Context-plugin and dynamic tools

Custom context plugins can provide instructions, content, tools, persisted state, compaction, and an IStoreHandler. Custom tool execution plugins do not add tools; they wrap execution with logging, analytics, argument transforms, caching, UI updates, or other policies.

Discovery APIs

import {
  ConnectorTools,
  ToolRegistry,
  ToolCatalogRegistry,
  listVendors,
} from 'oneringai';

// The 50 configured auth templates.
const templates = listVendors();

// Generic + specialized tools for one configured connector.
const githubTools = ConnectorTools.for('work-github');

// Every configured external connector that currently resolves to tools.
const byConnector = ConnectorTools.discoverAll();

// Service types with a registered specialized factory.
const specializedServices = ConnectorTools.listSupportedServices();

// Generated built-ins only, connector tools only, or their runtime union.
const builtIns = ToolRegistry.getBuiltInTools();
const connectorTools = ToolRegistry.getAllConnectorTools();
const allVisibleTools = ToolRegistry.getAllTools();

// Resolve scoped catalog categories for an application/tool picker.
const resolved = ToolCatalogRegistry.resolveTools(/* category scope */);

ConnectorTools.discoverAll() and ToolRegistry.getAllConnectorTools() inspect configured connector instances, not every possible template. A template does not become a live connector until the application creates or restores one.

Identity, scoping, and permissions

• Use identities on an agent to declare which connector names/accounts are visible to that agent.
• Use Connector.scoped() with an IConnectorAccessPolicy for tenant- or user-specific registry views.
• Use accountId for multiple accounts under one named connector.
• Use actAs only for the Microsoft/Google URL-scoped tools that support the on-behalf-of lock; tenant-global search/Drive/free-busy endpoints follow the underlying token's visibility.
• Connector tools still pass through ToolManager, so the normal permission policy chain, user rules, approval scopes, circuit breakers, and execution plugins apply.

Related documentation

• Connectors & Authentication
• Tools & Function Calling
• External API Integration
• Vendor Templates
• Tool Catalog
• Tool Permissions

Execution outcomes and cancellation

The shared ToolManager exposes additive executeWithOutcome() metadata and per-execution signals. Generic authenticated API tools forward ToolContext.signal to Connector.fetch(). Caller cancellation is composed with connector deadlines and remains effective during native response-body reading and retry backoff. Specialized tools and transports must forward/observe signals themselves; local cancellation does not guarantee remote effect prevention.

Connector header-timeout and retry defaults remain unchanged. Optional timeoutScope: 'operation' includes token waiting, retries, and body consumption; optional retryPolicy: 'safe' permits retries only for eligible GET/HEAD/OPTIONS. skipRetry skips network and status retries. Consume or cancel response bodies. See Tool outcomes and cancellation for outcome provenance, uncertainty, raw-result compatibility, and examples.