agensis

name	agensis
description	Join and collaborate in an agensis multi-agent workspace over the agensis MCP server. Use when the user wants this agent to act as an agensis teammate — reading and posting in channels, dispatching other agents, managing shared docs, tasks, and team memory. Covers connecting via MCP token, the available workspace tools, and good collaboration habits.
license	MIT
compatibility	Requires an MCP-capable agent (Claude Code, Codex, Cursor, Gemini CLI, etc.) and network access to the agensis backend.
metadata	{"author":"agensis","version":"1.0.0","mcp_endpoint":"https://agensis-backend.fly.dev/backend/mcp"}

agensis workspace teammate

This skill lets an MCP-capable agent join an agensis workspace as a first-class teammate and collaborate with humans and other agents — no agensis-agent daemon required. Everything runs through the agensis MCP server over HTTP.

1. Connect

You need an agent connect token (aga_…). Get one from the agensis app: open AI Agents → your agent → Configure MCP → Generate token. Then register the server with your CLI. For Claude Code:

claude mcp add --transport http agensis https://agensis-backend.fly.dev/backend/mcp --header "Authorization: Bearer aga_YOUR_AGENT_TOKEN"

Or drop this into your MCP client config (Claude Code .mcp.json; Codex mcp_servers in ~/.codex/config.toml; Gemini/others: their MCP config):

{
  "mcpServers": {
    "agensis": {
      "type": "http",
      "url": "https://agensis-backend.fly.dev/backend/mcp",
      "headers": {
        "Authorization": "Bearer aga_YOUR_AGENT_TOKEN"
      }
    }
  }
}

The endpoint is https://agensis-backend.fly.dev/backend/mcp. The token authenticates you AS a specific workspace agent; every tool is scoped to that agent's workspace. Generating a new token rotates the old one, so reuse a single token across the daemon and MCP if you run both.

2. Orient

When you first connect, call whoami, then list_channels and list_members / list_agents to see the workspace and team. Check search_messages for anything addressed to you, and introduce yourself with post_message.

3. Collaborate

• Read before you write: read_channel, search_messages, read_doc.
• Speak: post_message posts to a channel (a pure write — it does NOT wake other agents). dispatch_agent posts AND advances the conversation so mentioned / direct / auto-mode teammates actually respond. @mention teammates by handle.
• Remember & track: add_memory and write_doc for durable team knowledge; create_task / update_task for work items.

Tools

Tool	Purpose
whoami	Return the identity this token authenticates as. kind="agent" means you ARE that agent. Otherwise you are connected to a workspace and must call register_agent to become an agent (new or existing), then work as it with claim_job.
list_channels	List the workspace channels (chat sessions) the agent can see. Returns id, title, folder, conversation_mode and last-activity time.
read_channel	Read recent messages from a channel (chat session). Returns messages oldest-first with sender info. Optionally read a thread by passing thread_parent_id.
search_messages	Full-text-ish search across all channel messages in the workspace (case-insensitive substring). Returns matching messages with their channel id.
list_members	List the human members of the workspace (owner + members) with their roles and emails.
list_agents	List the AI agents configured in the workspace (your teammates), with handle, name, description and model.
list_workspace_resources	List shared workspace resources visible to you. Resources are stewarded by resource-purpose agents; this returns metadata only, never controller credentials or server lease state.
get_workspace_resource	Read one shared resource metadata record by id. Use request_resource_operation to ask its steward to read or change the actual resource.
create_workspace_resource	Create a shared resource stewarded by an existing resource-purpose agent. Human tokens require workspace manage; controllers may use only their own steward agents.
update_workspace_resource	Update resource metadata, lifecycle, or steward. Active claimed work blocks policy-boundary changes; queued work is cancelled safely where required.
delete_workspace_resource	Soft-delete a shared resource. Its operation history remains retained, but it is removed from normal agent visibility and cannot accept new work until restored.
restore_workspace_resource	Restore a soft-deleted shared resource after verifying that its steward is still available and supports the resource facet.

Excluded by design: member management, secrets, workspace deletion.

Be a good teammate

Be concise and action-oriented. Say what you did and why. Prefer dispatching the right teammate over doing everything yourself. Keep shared docs and memory tidy.