Skip to main content
Outlit is the product that monitors customers and completes approved customer work — its own opinionated agents produce the churn and renewal attention items your team reviews (including billing-risk evidence), and surface expansion insights inside customer context. The CLI, MCP server, and API let your coding agents and applications read authorized customer context and perform supported workspace actions.

​
Choose your interface

CLI, MCP, and the gateway share one underlying tool catalog and authorization model — but names and input shapes differ per surface. The CLI wraps tools in kebab-case commands/flags (outlit customers list --billing-status PAYING), while MCP and the API use tool names and camelCase schema fields (outlit_list_customers, billingStatus). Pick whichever host fits; capabilities and permissions are the same.

​
How an agent answers a customer question

This is the evidence-first pattern the interfaces are designed around:
1

Discover

outlit schema lists analytics views and columns. outlit customers list / outlit users list filter by billing status, journey stage, activity recency, and traits.
2

Get context

outlit customers get <customer> --include users,revenue,recentTimeline returns a bounded profile. outlit customers timeline filters activity by channel and event type.
3

Find evidence

outlit facts list returns structured facts with provenance (sourceType, sourceId, sourceOccurredAt, sourceQuote, permalink when available). outlit search "<natural language>" returns grouped artifact-level source and fact results — not raw vector chunks.
4

Open the source

outlit sources get --source-type CALL --source-id <id> returns the exact record. When a source offers contentPage, check contentPage.completeness before treating the content as complete — partial pages mean more content exists beyond what was returned.
5

Answer with evidence

Cite the sources and facts you actually retrieved. An absent permalink or empty result set is unavailable evidence, not negative evidence — say so instead of inferring.

​
Read vs. write controls

Most of the catalog is read-only customer intelligence. Writes are bounded and each requires an explicit grant: Outlit manages its monitoring rules and schedules. Public tools do not configure agents or signals, store or reveal credentials, disconnect integrations, or send arbitrary notifications.

​
Authorization model

  • API keys are independent workspace principals. A key’s grants are its own; they do not inherit the creator’s human permissions or customer access.
  • OAuth (MCP) runs as the signed-in member, so that member’s workspace permissions apply directly.
  • Read presets ≠ write presets. Read-only keys get discovery but not mutation; workspace_members:read and customer_access:manage are separate grants, and the discover-then-mutate flow needs both.
  • CLI output is script-friendly. See Output behavior for the exact JSON-mode rules.

​
Distinguish the MCP servers

  • Product MCP — https://mcp.outlit.ai/w/<workspace-slug>/mcp: customer intelligence tools described here.
  • Docs search MCP — the documentation-site search endpoint Mintlify hosts for this docs site. It searches these docs only; it has no customer data and no product tools.
Do not point customer-intelligence workloads at the docs endpoint, and do not expect product tools there.

​
Next steps

Customer workflows

Copy-paste task recipes: churn review, renewal prep, expansion, evidence trails

Troubleshooting

Auth, permissions, empty results, identity, JSON, MCP, and SDK delivery issues

Tool gateway API

The shared HTTP contract behind CLI and MCP

CLI commands

Every command, flag, and output schema