Run support with Claude Code

Murphy is agent-native on both sides. An AI agent answers your customers, and YOUR coding agent can work the inbox: pull a case, investigate against your codebase and data, reply, and resolve.

1. Get a workspace API key

Dashboard, Integrations, Generate API key. The key is scoped to one workspace.

2. Add the MCP server

claude mcp add murphy \
  -e MURPHY_URL=https://your-instance.com \
  -e MURPHY_API_KEY=ef_... \
  -- npx @askmurphy/mcp

Three tools: list_open_tickets, get_case_bundle (full transcript, customer identity, channel, pending drafts), and resolve_ticket (reply to the customer, log a resolution note, archive). Email tickets send real email; replies supersede any pending auto-draft.

3. Or use the REST API directly

# list open tickets
curl -H "Authorization: Bearer ef_..." \
  https://your-instance.com/api/v1/tickets

# full case bundle for one ticket
curl -H "Authorization: Bearer ef_..." \
  "https://your-instance.com/api/v1/case?id=42"

# reply, log the resolution, close the ticket
curl -X POST -H "Authorization: Bearer ef_..." \
  -H "Content-Type: application/json" \
  -d '{"id":42,"reply":"...","resolutionNote":"...","resolve":true}' \
  https://your-instance.com/api/v1/resolve

Teach the agent from your codebase

Your code knows things your website does not: exact limits, error messages, edge-case behavior. Run this in your repo (once, or as a weekly scheduled agent). Your code never leaves your machine — only the support-safe summary does.

> Read this codebase and distill a support knowledge document:
  every user-facing feature and how it behaves, limits and quotas,
  error messages a user might see and what they mean, pricing
  constants, and common gotchas. Support-safe prose only — no
  source code, no secrets. Then call sync_knowledge with slug
  "codebase-knowledge".

Ground rules for agents

Never tell a customer an action happened (refund, cancellation) unless it actually executed in the relevant system.
Email replies are plain text. No markdown, no em dashes.
Log a resolution note on every ticket you act on, so the humans in the inbox see what happened and why.

Machine-readable overview at /llms.txt.