Skip to main content
The MCP server implements the Model Context Protocol so AI tools (Cursor, Claude Code, Copilot, custom hosts) call first-class tools instead of scraping free text. Marketing copy describes it as giving your stack a persistent brain across every conversation and IDE; technically, it is a hosted MCP endpoint that authenticates the user and proxies to Hopsule’s API.

​
What it does

  • Exposes structured tools grouped by domain: navigation (orgs, projects, active project), decisions, memories, tasks, context (capsule- and project-scoped retrieval), and status helpers.
  • Enforces role-based tool visibility: membership role in an organization determines which tool names are registered for that session.
  • Requires a coherent session project: navigation tools instruct agents to set an active project before project-scoped calls, or to pass an explicit project identifier where supported.
Typical agent flows mirror the landing page demos, for example list_decisions, create_memory, get_context-style operations with low-latency round trips to your live project state.

​
When to use MCP vs other surfaces

Use MCP when…Prefer IDE add-on when…Prefer CLI when…
The client is an LLM agent or chat UI with MCP supportYou want diagnostics in the editor and sidebar browsingYou want a TUI, scripts, or CI push/status
The IDE add-on can run MCP alongside local editing; both ultimately respect the same ACCEPTED decisions from the API.

​
Security and auth

The hosted server integrates with Clerk-compatible auth for MCP (including OAuth flows appropriate to MCP clients). Treat MCP credentials like any production API access: least privilege via org roles, no embedding long-lived secrets in prompts.

Platform overview

How MCP fits the full product map.

IDE add-on

Editor-native MCP and enforcement.

Memories

What agents retrieve and cite.

Decisions

Immutable rules agents must not override.