MCP Overview
anymrkt exposes a single MCP (Model Context Protocol) server that unifies every product into one surface an AI agent can connect to. The server runs as an HTTP endpoint at /mcp, authenticated with a bearer token. Any MCP-compliant client can connect.
The three layers
1. MCP Server — what anymrkt runs today
Built on @modelcontextprotocol/sdk, our server registers 119 tools across 7 product categories and serves them over HTTP. The server is per-request: each connection resolves a ResolvedMcpToken from the bearer header, and the tool list is filtered based on that token's permissions and optional tool_filters allow-list.
| Category | Tools | Products |
|---|---|---|
| Listings & brand | 14 | Inventory, conversations, brand content |
| Builder | 6 | Page editor — sections, templates, layout |
| Scheduler | 9 | Meeting types, availability, bookings |
| GRAIL | 3 | Property records, GIS signals, imagery |
| MACP | 38 | Boards, tasks, sessions, connectors, webhooks |
| Kanban | 29 | Kanban boards, cards, journeys, sprints, epics |
| Studio | 20 | API keys, connectors, usage analytics |
2. MCP Client — agents connecting to anymrkt
This is the direction you use today. Configure your agent with our HTTP endpoint and a bearer token.
See the transport guide for per-client setup.
3. MCP App — interactive UI inside agents (planned)
MCP 2026 introduced MCP Apps: interactive applications that render UI inside the AI client. anmrkt does not yet ship an MCP App, but the architecture is ready. Potential apps:
- Builder page preview inside Claude Code — edit, see a live render, confirm
- Calendar availability grid — click to toggle open/closed blocks
- Kanban board view — drag cards, the agent proposes moves, you confirm
- GRAIL map with signal markers — click a pin for parcel details
Security model at a glance
| Layer | Mechanism | Prevents |
|---|---|---|
| Transport | DPoP optional binding | Stolen bearer token is useless without the client key |
| Auth | Per-token permissions + brand scoping | Token for brand A can't touch brand B |
| Writes | Permission-gated | Agent can't act outside granted permissions |
| Public actions | confirm: true required | Can't happen as a side effect |
| All mutations | preview: true dry-run | Agent shows what it will change first |
| Audit | mcp_operations log + revert_run | Full revert per agent session |
| Tool filtering | Per-token tool_filters | Agent only sees tools you enabled |