Claude Code plugin

plugin/ is a Claude Code plugin that lets you ask the daimons in your Slack workspaces and Discord servers a question from inside Claude Code. Teams has no login mount yet, so the plugin can't reach daimons on Teams. plugin/README.md is the user-facing doc: what the plugin ships, how to install it, and how to log in. This page is the part you want before you install — what has to be true on the server side for any of it to work.

What it is

The plugin declares two HTTP MCP servers, daimon-slack and daimon-discord, and carries no credentials of its own. It adds one skill, daimon-context, which routes a question about team context to the right daimons and merges their answers, and one command, /daimon:daimon-status, which lists what is connected without asking anything. Both are built from the eight tools registered by packages/adapters/mcp/daimon/adapters/mcp/tools/hub.py, catalogued in mcp-tools.md — a surface separate from the main MCP tool set.

The hub login mounts

Those two URLs only exist if the deployment you point at has the hub login mounts configured, and "mount" here means a URL path, not a file or a volume. packages/adapters/mcp/daimon/adapters/mcp/hub/app.py builds one FastMCP sub-app per platform and mounts it on the MCP server's ASGI app at /slack and /discord, each with its own OAuth proxy, so a Slack login token is never evaluated by the Discord proxy. The plugin then talks to <origin>/slack/mcp and <origin>/discord/mcp.

A platform's mount appears only when both that platform's client id and secret are set, so a deployment can offer one, both or neither, and one that sets neither behaves exactly as it did before the mounts existed. Beyond the DAIMON_HUB__* pair, the mounts need DAIMON_CRYPTO__KEYS (login state is encrypted at rest) and DAIMON_MCP__PUBLIC_URL (they derive their public base URL from it); the server refuses to boot if a mount is configured without them. The operator side — OAuth apps, redirect URIs, the scopes each platform asks for — is in self-hosting.md, and the settings themselves are in configuration.md.

What a login gets you

Logging in proves one platform identity and the workspaces it belongs to. packages/core/daimon/core/hub_identity.py intersects that set with the tenants where daimon is actually installed and ready, and looks up or provisions your account in each one exactly as the chat adapters do on first contact — so a turn you start from Claude Code is permission-checked and billed as you, in that workspace, not as some shared service identity. The intersection is recomputed on every call, and Slack reads run with your own Slack visibility.

That also means the answer to "why do I see no daimons" is usually not the plugin. Either you have not authenticated that server with /mcp yet, or daimon is not installed in any workspace you belong to, or the deployment has no mount for that platform.

Before you install

• Point it somewhere. It defaults to the hosted deployment. For your own, set DAIMON_MCP_URL to the origin — scheme and host only, since the plugin appends the platform paths itself.
• Log in per server. Installing the plugin authenticates nothing; you run /mcp and authenticate each platform you want.
• Know the billing behaviour. A turn started through the hub runs the same balance and cap admission checks as any other MCP turn, so a workspace out of credit refuses it up front. As with those tools today, it does not record a metered debit when it completes. See billing.md.
• Installing from a marketplace is not available yet. Until the plugin is published, the working path is claude --plugin-dir ./plugin from a checkout, as plugin/README.md describes.

One operational note for self-hosters: the hub keeps short-lived login state in a table that is pruned by a sweep on the scheduler's tick (packages/core/daimon/core/hub_oauth_kv_sweep.py). The store honours expiry on read but never deletes, so a deployment that runs the mounts without a scheduler grows that table by a row per login attempt.