Saya plugins

Saya plugins

Plug any agent into your team's Saya.

Before you guess, ask the team brain. Saya exposes a central remote MCP endpoint so external agents can query live, trust-graded team knowledge from the workspace Saya already lives in.

saya.computer · MCP: https://saya-mcp.luke-nittmann.workers.dev/mcp

Install

Any MCP client (.mcp.json)

Add Saya as a Streamable HTTP MCP server at the live workers.dev origin. When your client prompts, complete browser OAuth for your team; the per-team bearer belongs in the client credential store, never in this repo.

{
	"mcpServers": {
		"saya": {
			"url": "https://saya-mcp.luke-nittmann.workers.dev/mcp",
			"transport": "streamable-http"
		}
	}
}

Any agent (npx skills)

Installs the secondary usage-guide skill for skill-aware agents. This teaches when to query Saya; the live MCP endpoint remains the product surface.

npx skills add creative-int/saya-plugins

Claude Code

Add the marketplace, install the Saya plugin, then complete browser OAuth before the first team-brain query.

/plugin marketplace add creative-int/saya-plugins
/plugin install saya@saya

Codex

Add this repo as a Codex plugin marketplace, install from /plugins, then authenticate the MCP connection for your workspace.

codex plugin marketplace add creative-int/saya-plugins

Cursor

Install Saya from the Cursor plugin marketplace, then authenticate the MCP connection for your workspace.

Cursor -> Settings -> Plugins -> Add marketplace -> creative-int/saya-plugins

To preview the available skills without installing:

npx skills add creative-int/saya-plugins --list

Quickstart: tap the team brain

1. Add the MCP server with the generated .mcp.json block above, or install theplugin through Claude Code, Codex, or Cursor.
2. When the client prompts for auth, complete browser OAuth with the workspacethat owns the Saya brain. The bearer is per-team and should stay in the MCPclient's credential store or environment, never in a tracked file.
3. Start with saya_status if you need to confirm the live bridge, scopes, orworkspace health.
4. Use saya_context before making team-specific assumptions. Ask focusedquestions like: "What has this team decided about the Saya MCP launch? Returnverified decisions first and call out candidate or deprecated notes."
5. Use saya_act only for the curated approval-first memory save path. Treat itas a write that needs user intent, not a general automation channel.

If a client exposes raw MCP calls instead of a friendly tool picker, inspecttools/list after OAuth and follow the current input schema advertised by theserver.

MCP surface

Saya exposes one central Streamable HTTP MCP server. Team identity and accesscome from per-team OAuth bearer auth plus workspace membership; never put tokensin repo files.

Status / readiness

The deployed endpoint is live on prod athttps://saya-mcp.luke-nittmann.workers.dev/mcp. Production OAuth andworkspace membership are required; authenticated saya_context queries returnreal, trust-graded team knowledge from Saya's Convex brain. saya_act is theapproval-first save_memory path, and saya_status reports live bridge health.

The custom domain https://mcp.saya.computer/mcp is not routed yet, so thisrepo intentionally points at the current workers.dev origin.

Included skill

• saya-team-brain — secondary usage guidance for agents deciding when toquery Saya through MCP before answering. Useful for team memory, decisionhistory, workspace norms, channel context, and "what does this team alreadyknow?" questions. Also answers to ask-team and team-context.

The skill lives atskills/saya-team-brain/SKILL.md.

See docs/team-brain-agent-tap.md for therunner-side sample flow and proof boundary.

Develop

pnpm install
pnpm generate        # regenerate all adapters from saya.config.ts
pnpm verify          # drift check + typecheck + build/help + MCP smoke

pnpm smoke always checks unauthenticated protected-resource metadata. SetSAYA_MCP_BEARER to additionally assert the authenticated tools/list surface.

License

MIT © creative-int