Skip to content
Vonzio docs

vonzio (platform) MCP server

The vonzio MCP server (URL: /mcp/platform) is how agents call back into the platform itself. Use cases:

  • A coordinator agent submits work to a worker profile
  • A playbook agent triggers another playbook on completion
  • An agent queries memory across profiles for cross-domain context

Always injected (when INTERNAL_SERVER_URL is set). Authorization is bound to the running user — these are “operate as me” capabilities, not “operate as someone else”.

Tools

Section titled “Tools”

task_submit

Section titled “task_submit”

Submit a new task to any profile owned by the running user. Useful for fan-out.

{
  "profile_id": "prof_research_assistant",
  "prompt": "Find me everything we have on quarterly revenue trends.",
  "mode": "batch",
  "max_turns": 50
}

Returns the new task_id. The submitting agent doesn’t block on completion — it’s fire-and-forget.

task_wait

Section titled “task_wait”

Pair with task_submit: block until a submitted task completes, then return its result text.

{ "task_id": "task_…", "timeout_seconds": 120 }

Returns either the assistant’s final text or an error if the task failed / timed out.

playbook_run

Section titled “playbook_run”

Manually fire a playbook owned by the running user.

{ "playbook_id": "pb_…" }

Returns the new run_id. Useful for chaining: “Run the cleanup playbook, then run the report playbook.”

workspace_list

Section titled “workspace_list”

List the user’s workspaces. Same shape as GET /v1/workspaces but available to the agent without the agent holding an API token.

{ "status": "active", "limit": 20 }

workspace_archive

Section titled “workspace_archive”

Archive a workspace (sets archived=true). Use for hygiene playbooks that clean up stale conversations.

{ "session_id": "sess_…" }

memory_query_global

Section titled “memory_query_global”

Query memory across all the user’s profiles (not just the running one). Useful for coordinator agents.

{ "query": "credit card payoff plan" }

The running agent still can’t write to other profiles’ memory — only the running profile’s. Use memory_write on the memory MCP for that.

Authentication

Section titled “Authentication”

Authorization: Bearer platform_<token>. The bearer carries { userId, profileId }. All operations respect user ownership — no cross-user reach.

Security notes

Section titled “Security notes”
  • Agents can’t escalate privilege via this server. They can only operate within their user’s existing permissions.
  • Cross-profile reach is limited to memory_query_global (read-only across profiles). Mutations remain per-profile.
  • playbook_run can fire any of the user’s playbooks. If you don’t want an agent to be able to trigger a sensitive playbook, scope it to a different profile (a profile the running agent isn’t on).