Docs · MCP

Your company brain, for any agent.

Clearly is an MCP server — not a plugin. Agents connect out to it. Once connected, Claude Code / Cursor / any MCP client can search your org’s context, write decisions back so they compound, schedule their own follow-ups, and inherit your team’s skills. The server is hosted at relay.clearly.sh/mcp — nothing to deploy or run.

Connect an agent

Sign in through the browser, then point a client at the hosted endpoint. No token to mint or paste.

1 · Sign in over OAuth

OAuth browser sign-in on first connect — nothing to mint or paste. Scopes are granted at sign-in; revoke anytime with claude mcp logout clearly.

https://relay.clearly.sh/mcp
rpc:read   ·   rpc:write   ·   admin

rpc:read — search + read. rpc:write — also create documents + generate art. admin — owner-only.

CLEARLY_MCP_URL=https://bee-relay-staging.throbbing-unit-1b3f.workers.dev/mcp

Point at the staging relay before launching the client.

Want the connection to act as a specific agent (gated by that agent’s access + attributed in activity)? Start the sign-in from the agent’s card in the Agents modal → Connect over MCP. See Agent identity below.

2 · Install

Claude Code: the plugin wires the server AND ships the skills. Any other client: drop the endpoint into its config, then complete the browser sign-in.

/plugin marketplace add clearly-sh/clearly-plugin
/plugin install clearly@clearly
# then: /mcp → select clearly → Authenticate (opens browser sign-in)
claude mcp add --transport http clearly https://relay.clearly.sh/mcp
claude mcp login clearly   # opens the browser for OAuth sign-in
{
  "mcpServers": {
    "clearly": {
      "url": "https://relay.clearly.sh/mcp"
    }
  }
}

~/.cursor/mcp.json or claude_desktop_config.json — no headers; OAuth-capable clients prompt a sign-in

What your agents gain

The headline tools — plus the full ~200-action workspace surface via the catalog. The canvas has its own guide: /docs/canvas-api.

Company brain
  • clearly_context_search— One ranked search across prompts, docs, decisions + facts — this workspace, or the whole org with scope:"org".
  • clearly_context_write— Write a PRD / decision back so the next agent finds it (private:true keeps it workspace-local).
  • clearly_context_map— Orient: brain stats, recurring topics, and the orgs it can federate across.
Skills
  • clearly_skill_list— Discover the workspace’s reusable procedures.
  • clearly_skill_get— Load one skill’s full step-by-step instructions on demand.
Project board
  • clearly_ticket_create— Open a card on the project board — a card is a ticket (the one board model; /projects/board + the conversation Tasks tab render these).
  • clearly_ticket_update— Move a card (set its status) or edit it — priority, assignee, due date, tags.
  • clearly_tickets_list— List a project’s cards, optionally filtered by status.
The canvas
  • clearly_canvas_perceive— SEE a composition — every node, its box + style, the selection, the viewport. Optionally a rendered PNG.
  • clearly_canvas_act— BUILD on it — frames, text, vectors, images, auto-layout, components. One action or a whole batch. Works headless.
  • clearly_canvas_catalog— Every canvas verb + its exact args, and which ones run without a browser tab. Generated, so never stale.
Guidance
  • clearly_guide— The Company Brain usage playbook — call it once on connect to learn how to use the rest.
Everything else
  • clearly_workspace_catalog— List all ~200 workspace actions, with schemas + examples.
  • clearly_workspace_invoke— Run any workspace action by name.

Agent identity & sessions

Make an MCP connection behave like the desktop daemon — an agent acting as itself.

Agent-scoped sign-in

Start the sign-in from the Agents modal → Connect over MCP.

A normal sign-in acts as you. An agent-scoped connection acts as one of your agents: every call is gated by that agent’s own scope / tool-allowlist (a read-only agent can’t mutate even though you can) and attributed to it in the activity log.

Sessions

Streamable-HTTP transport.

initialize issues an Mcp-Session-Id; clients echo it on subsequent calls so a session’s tool calls correlate. Stateless if absent — every client keeps working unchanged.

The usage playbook

How to actually work the brain — available to every client, no plugin.

Call clearly_guide (or the clearly-workflows MCP prompt, surfaced as a slash command in Claude Code) to get the loop:

1. context_map        — orient
2. context_search     — read before you write (scope:"org" to federate)
3. context_write      — file decisions back so they compound
4. skill_list/get     — follow the team's procedures, don't improvise

Rules of thumb: read → act → write back; scope:"org" for cross-team questions; private:true for sensitive writes.

Skills your agents inherit

Claude-style procedures — a name, a when-to-use trigger, a step-by-step body.

A connected agent calls clearly_skill_list to see your team’s skills, then clearly_skill_getto load one when its trigger matches — so it works the way your workspace works, not from a blank slate.

Clearly ships no default skills — the library is grown: authored by hand, generated when you create an agent, or researched autonomously by a skill-research scheduled run (web-grounded, with sources). It compounds as your team works.

From the terminal (CLI)

The beehaven CLI drives the same workspace from your shell (needs the Hive desktop app running).

beehaven connect home
beehaven actions                                    # everything you can call
beehaven call context-search '{"query":"roadmap"}'  # search your brain
beehaven call skill-list '{}'                       # your team's procedures
← Beehaven CLIAll docs →