Tools vs. Connectors
Tools and Connectors both give your agent mid-call capabilities — here's the real difference between them and when to reach for each.
Talkr has two separate places to give an agent a capability: Tools (/tools) and Connectors (/connectors). They both end up as things the LLM can call during a conversation, which is why they're easy to mix up — but they solve different problems.
The short version: a Tool is something you build — an arbitrary HTTP request or an MCP server connection, wired up with your own URL, headers, and credentials. A Connector is something Talkr already built — a named integration with a specific third-party service (Salesforce, HubSpot, Slack, Razorpay, and others) that ships with ready-made actions.
What a Tool is
A Tool is a capability you define yourself, from scratch. Talkr supports two kinds:
- HTTP API — you supply a URL, method, headers, authentication, and parameters, and the LLM can call that exact endpoint during a call. This is how you reach any system: your own backend, an internal API, or a third party Talkr doesn't have a Connector for. See HTTP API.
- MCP Server — you point Talkr at a remote Model Context Protocol server URL (plus an optional credential), and Talkr discovers and exposes whatever tools that server defines. See MCP Tool.
Both are created from the Tools page, and both require you to know the endpoint or server you're connecting to — Talkr doesn't know anything about the service on the other end beyond what you tell it.
What a Connector is
A Connector is a pre-built, Talkr-maintained integration with one specific, named service. Talkr already knows the API shape, the available actions, and (for most providers) how to complete OAuth against it. You just bring your own account.
- Connect once per organization on the Connectors page, using your own account's credentials (OAuth authorization, an API key, or — for a few caller-ID/trust providers — a manual request-access form).
- Once connected, the connector's actions (e.g. "Create Lead," "Post Message," "Look Up Order") appear automatically in every agent's tool picker for your organization. There's no separate step to build the request.
See Connectors Overview for the full catalog and how authentication works.
When to use which
| Tool | Connector | |
|---|---|---|
| The service | Any URL, or any MCP server | A specific service Talkr has built-in support for |
| Who defines the request | You (method, headers, body, parameters) | Talkr (pre-built actions) |
| Setup effort | Build the HTTP call or point at an MCP server | Authorize your account, no request-building |
| Scope | Per tool, attached to nodes individually | Org-wide once connected; actions available to every agent |
| Best for | Internal systems, niche APIs, anything custom | Well-known CRMs, helpdesks, messaging, commerce, payments |
A simple way to decide: check the Connectors catalog first. If the service you need is listed there, connecting it is almost always less work than building the equivalent HTTP tool by hand — you skip writing the request shape and, for OAuth providers, the credential handling too. Only reach for a Tool when the service isn't in the catalog, or when you need something a pre-built action doesn't cover (a custom internal endpoint, a one-off automation, a self-hosted MCP server).
Examples
Tool
- Call your company's internal inventory API to check stock for a SKU the caller asks about — no Connector exists for your internal systems, so this has to be an HTTP API tool.
- Trigger a self-hosted MCP server that exposes your order-management functions to the LLM.
- Hit a niche SaaS product's REST API that Talkr doesn't have a Connector for, using an API key you paste into the tool's headers.
Connector
- Create a Salesforce lead when a caller expresses interest — connect Salesforce once via OAuth, and "Create Lead" is immediately available to attach to a node.
- Post a message to a Slack channel when a call needs human follow-up.
- Create a Razorpay payment link and read it back to the caller.
Tools and Connectors aren't mutually exclusive on the same agent — a single node can have an HTTP API tool, an MCP tool, and a Connector action attached at once. The LLM picks whichever one fits the caller's intent based on the descriptions and your node prompt.