# Usage Guide

Learn how to wire the Outreach MCP server into your agent runtime and build effective workflows.

## 1. Wire the MCP server into your agent runtime

Most agent frameworks (Claude, MCP clients, custom orchestrators) accept an MCP server URL plus auth. Once connected, the tools above are auto-discovered via `tools/list`. You don't list them manually.

## 2. Let the agent pick the tool

Tools are described with schemas the model reads at runtime. Your job is the *prompt*, not the API call. Examples:

| User asks... | Agent calls... |
|  --- | --- |
| "Find CROs at Contoso" | `prospect_search` with title filter |
| "What's the status of the Acme deal?" | `opportunity_search` → `opportunity_get_by_id` |
| "Create a prospect for Sarah Chen at Geometric Corp" | `prospect_create` |
| "Add her to the outbound sequence" | `sequence_search` → `sequence_add_prospects` |
| "Show me recent meetings with Initech" | `kaia_meeting_search` with account filter |


## 3. Chain tools for multi-step workflows

The agent does this automatically. For "Add Sarah Chen to Executive Outbound Touch":

1. `prospect_create` → returns prospect ID
2. `sequence_search` (name = "Executive Outbound Touch") → returns sequence ID
3. `sequence_add_prospects` (both IDs) → returns batch status


No glue code on the developer side.

## 4. Handle write operations carefully

Writes are real. Build in confirmation flows, audit logging, and rate limits in your agent layer. The MCP server does not throttle or undo. Use `destructiveHint: true` annotations to drive your confirmation UX.

## Common Tool Patterns

### Search → Fetch → Act


```
account_search → account_get_by_id → account_answer_question
```

Most exploration workflows follow this shape.

### Create → Enroll → Verify


```
prospect_create → sequence_add_prospects → sequence_state_search
```

The full prospect onboarding lifecycle.

### Discover Schema → Create


```
input_fields_fetch → prospect_create
```

Use when fields differ across orgs (custom fields, required validations).