AGENT ATLASa field guide to agent workflows
Back to Memory & context
Memory & context

Tool connectors

Wire the agent into your real tools and data sources.

N agentsCore

How it works

  1. 1Connect the agent to each system through a defined interface — an MCP server or API client.
  2. 2Expose only the specific capabilities the task needs (least privilege).
  3. 3Pass credentials securely, never in the prompt.
  4. 4Test each connector's actions in isolation before chaining them.

Use it when

When the agent needs to act in systems you already use — calendar, docs, repos, CRM — through a standard interface rather than bespoke glue.

Reach for something else when

A single simple API call suffices. A full connector layer is overkill for one endpoint.

Where you stay in the loop

You grant least-privilege access and flag state-changing actions for approval; the agent acts through the connectors. Credentials and write-access are yours to govern — test each connector in isolation before trusting it.

In the wild

An agent reads your calendar and docs through connectors to prepare a meeting brief.

Hand this to your agent

Help me plan the connectors an agent needs for my workflow.

List: (1) each external system it must read from or write to, (2) the minimal
set of actions to expose for each (least privilege), (3) which actions are
read-only vs state-changing — flag the state-changing ones for approval, (4)
what to test per connector before going live.

Workflow: <...>

Replace the <…> placeholders, paste it into your agent, and it'll scaffold the workflow with you.