> ## Documentation Index
> Fetch the complete documentation index at: https://docs.linkedcamp.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Scope and safety

> What the MCP server can and cannot do, which tools contact real people, and the built-in prompts and resources.

## Available

* Campaigns, sequences and flows
* Prospects, activities and notes
* The unified inbox — reading, AI-drafted replies and sending
* LinkedIn and email account settings
* Tags, blacklists, templates, merge fields
* Personas and AI agents
* LinkedIn posts and scheduling
* Analytics and reporting
* Tasks, appointments, reminders
* Account-wide automations

## Not available

These operations are not exposed by any tool, whatever the request says.

| Area | Why |
| - | - |
| Billing — subscriptions, plans, payment methods, invoices, coupons | Money movement is never delegated to an integration |
| Agency and sub-account administration, licences, impersonation | Account-owner actions with cross-tenant reach |
| Tenant / white-label configuration | Platform configuration, not account data |
| Team invitations, role changes, session revocation | Membership is an account-owner action |
| Account deletion, trial termination, cross-account campaign moves | Irreversible and out of an integration's remit |
| Reading back stored credentials — LinkedIn passwords, mailbox tokens, integration API keys | Excluded from every selection set; no tool returns them |
| Connecting a LinkedIn account, PIN verification, mailbox OAuth callbacks | Take passwords, PINs and OAuth codes; belong in a browser |
| CRM integrations — connecting, reading, field mapping, pushing a lead | Connecting one takes a third-party credential; configure integrations in the web app |
| Deleting a saved reply or a lead note, a sequence, an appointment or a notification | Irreversible record loss with no safer equivalent; campaign, holiday, reminder and the AI-library deletions are available with confirmation |
| Rotating the API key | Invalidates the key the connector is using, cutting its own connection mid-session |
| Archiving or completing a campaign | Web-app housekeeping with no outreach meaning; pause and resume are available with confirmation |

<Note>Exclusions are enforced on the server. An API key that asks for a blocked operation gets `403 FORBIDDEN` regardless of how the request is phrased — the check inspects the GraphQL root field being executed, not the label on the request.</Note>

## Permissions

Each tool group is tied to a section permission on your account (Campaigns, Accounts, Inbox, Reporting, Personas, Agents, Publish, Tasks, Automations). If your plan or team role does not grant a section, the server hides its tools from `tools/list` and refuses them on call. Pro-only sequence steps (`EMAIL`, `QUALIFIER`, `GPT`, `FINDER`, `EVENT`) are refused on plans without Pro features — call [`whoami`](/mcp/tools/core#whoami) to see what is locked.

## Destructive tools

These tools contact real people or delete records permanently. Each is marked **destructive** in the protocol, so your MCP client asks for confirmation before running it.

| Tool | What happens |
| - | - |
| `campaign_launch` | Outreach starts sending on the campaign schedule |
| `campaign_delete` | Moves the campaign to DELETED so it stops appearing — reversible in the database, not from here |
| `flow_set on an ACTIVE campaign` | Re-evaluates every lead against the new sequence and can cancel pending actions |
| `campaign_send_test_email` | Sends a real email |
| `post_upsert with a due scheduledAt` | Publishes to LinkedIn publicly |
| `message_send` | Delivers to a real person on LinkedIn or by email immediately — cannot be recalled |
| `lead_delete` | Removes leads and their pending actions permanently |
| `lead_blacklist` | Excludes someone from every future campaign on the account; nothing removes it |
| `message_delete` | Removes a message from your record only — the recipient keeps it, it is not unsent |
| `agent_delete` | Campaigns using it lose automated replies, so prospects mid-conversation stop getting answers |
| `kb_delete` | Source material is gone — a crawled page must be crawled again |
| `persona_delete` | Breaks every QUALIFIER step pointing at it; those flows then fail to save |
| `tag_delete` | Leads lose the classification and past reply-sentiment reporting changes |
| `blacklist_delete` | People the rule was blocking — including opt-outs — become targetable again |
| `placeholder_delete` | Copy using the merge field renders empty with no error, so live campaigns send with a hole in them |

<Warning>Editing a live campaign's sequence with `flow_set` re-evaluates every lead already in flight. Call [`flow_audit`](/mcp/tools/flows#flow_audit) straight afterwards to see exactly what changed.</Warning>

## Built-in prompts

The server publishes these MCP prompts. Clients that support prompts show them as slash commands or templates.

| Prompt | What it does |
| - | - |
| `launch-campaign` | Plans and launches a campaign, with a mandatory review step |
| `campaign-health-check` | Diagnoses an underperforming campaign and proposes fixes |
| `triage-inbox` | Classifies unread replies and drafts responses for approval |
| `weekly-report` | Cross-campaign performance summary with week-over-week movement |
| `clone-and-retarget` | Copies a working campaign onto a new audience |

## Resources

| URI | Title | Description |
| - | - | - |
| `mcp://reference/schema` | Platform vocabulary | Every enum the API accepts — call the reference tool or read this resource. |
| `mcp://guide/campaign-playbook` | Campaign building playbook | Full build-a-campaign walkthrough with a worked example. |
| `mcp://guide/safety` | Scope and safety | What the assistant will do only after asking, and what is never available. |


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.