Knowledge BaseKnowledge Base

Setting Up the Wazzi Agents Connector

Expose the agents you've built in Agent Center to any AI tool that speaks MCP — list, inspect, execute, monitor, and abort agent runs from outside the Wazzi dashboard. First-party, org-scoped, and requires no credentials.

August 12, 2026
6 min read

Path: Connectors → Wazzi Agents → Configuration

What this connector does

Wazzi Agents is a first-party MCP connector that exposes your organization's agents — the automations defined in Agent Center — to external AI tools. It lets an outside assistant (Claude, ChatGPT, Cursor, or another Wazzi agent) list the agents you've defined, inspect their configuration, kick off an execution, poll its status, abort a run, and browse execution history. Think of it as the "Wazzi API to Wazzi agents": one MCP surface that makes every agent in Agent Center invokable from anywhere you've connected the MCP.

Because Wazzi Agents is first-party, there's nothing to configure and no external credentials to enter. Every request is org-scoped server-side, and executions run under the same tokens your agents already use.

The connector exposes 9 tools:

  • Discovery (3 tools) — list_agents, list_agent_tags, get_agent to find and inspect agents defined in Agent Center.
  • Execution (2 tools) — execute_agent to kick off a run, abort_execution to cancel one mid-flight.
  • Monitoring (4 tools) — list_executions, get_execution_status (fast poll), get_execution (full detail), and get_action_batch_items (per-item results for batch actions).

Before you start

Wazzi Agents requires Agent Center to be enabled on your organization. Without it, the agent catalog is empty and execute_agent has nothing to invoke. If Agent Center isn't yet enabled, contact your Wazzi admin.

Beyond that, there's no per-user setup — just enable the connector on the user's group in Wazzi.

What this connector unlocks for your team

Once configured, members with the right MCP permissions can ask Claude / ChatGPT / Cursor things like:

  • "List every agent tagged 'onboarding' and kick off the welcome-email one for guest X."
  • "Run the 'utility fee estimator' agent for property Y and give me the output."
  • "Kick off the 'daily orange' agent for the Phoenix market and let me know when it's done."
  • "What's the status of execution <id> — is it still running?"
  • "Show me the last 20 executions of the 'call summary' agent and flag any that failed."
  • "Abort execution <id> — the guest replied on their own so we don't need the follow-up."
  • "Pull the per-item results from batch execution <id> — I want to see which units succeeded and which errored."

The exact set of available actions depends on which MCP toggles are flipped on for the user's group in Wazzi — see Managing Permissions. Note that which agents the connector exposes is also permission-scoped — a user only sees agents their group can execute.

Steps — enabling Wazzi Agents in Wazzi

1. From the Connectors catalog, click Configure on the Wazzi Agents tile.

You'll land on Wazzi Agents' Configuration tab:

Wazzi Agents connector Configuration tab

2. Read the notice — there are no credentials to enter.

The configuration tab shows an information notice reminding you that Agent Center must be active for this connector to have anything to run.

3. Click Test Connection.

Wazzi checks that (a) the agents catalog is reachable and (b) the agent-backend URL is configured server-side. On success you'll see:

  • "Connected. 12 agents available." — Agent Center is active and you have agents defined.
  • "Connected. No agents defined yet — create one in Agent Center to get started." — Agent Center is enabled but you haven't built any agents yet.

4. Click Save Configuration.

Wazzi Agents moves to the Active section of the catalog and the dot turns green.

Troubleshooting

  • Test Connection says "Agent executor URL not configured". The Wazzi backend needs AGENT_BACKEND_URL (or MCP_BACKEND_URL) set for execute_agent and abort_execution to route correctly. Contact your Wazzi admin — this is a server-side configuration, not something you set in the connector UI.
  • list_agents returns an empty list even though I see agents in Agent Center. The calling user's group doesn't have execute permission on any agents. Grant per-agent access in the agent's Access tab, or grant blanket agent access at the group level — see Managing Permissions.
  • execute_agent starts but never completes. Something is wrong with the agent's own configuration (missing credentials on a downstream connector it uses, an infinite HITL loop, etc.). Use get_execution to inspect the last-completed action and the current step — it points at what to fix.
  • "Not this user's turn" or permission errors on a specific agent. The agent is gated on per-agent permissions that the caller's group doesn't have. Ask an admin to grant execute access on that agent.
  • Executions poll forever with "queued" status. The agent-backend queue is backed up — usually due to browser-agent capacity limits during high load. Wait a few minutes and re-check get_execution_status. If it doesn't progress, contact Wazzi support.

Best practices

  • Poll cheaply with get_execution_status, not get_execution. The status tool returns just the state (queued / running / completed / failed / aborted); the full get_execution is heavier. Use full-detail only when you actually need the log / outputs.
  • Tag your agents. list_agent_tags + tag filters on list_agents lets callers narrow to the domain they care about (e.g. "show me all onboarding agents") — much friendlier than a flat list of every agent in the org.
  • Use abort_execution liberally. A stuck or unwanted execution consumes browser-agent capacity; aborting frees it up for other work.
  • Batch agents: use get_action_batch_items to see per-item outcomes. A batch agent that iterates over 200 properties returns one execution — but each iteration is an action batch item with its own success/error. This is the tool that answers "which items succeeded."
  • Never expose execute_agent to public / untrusted callers. Executions run on your infrastructure with your agents' full capability. Only grant this tool to internal groups whose usage you trust — see Managing Permissions.

Frequently asked questions

Do I need to configure anything?
No. This is a first-party connector with no external credentials. Just enable it on the user's group and (ensure Agent Center is active on the org).

Can users from another organization run my agents?
No. Every tool call filters by organization_id — the agents catalog, execution history, and execute path are all org-scoped. Cross-org access is impossible.

How is this different from just using the agents through the Wazzi dashboard?
The dashboard runs agents from the Wazzi UI. This connector lets external AI tools (Claude Desktop, ChatGPT, Cursor, or a chat-driven meta-agent) invoke them. Same agents, same execution engine — different launch surfaces.

Can I call one agent from another agent?
Yes — the Wazzi Agents MCP can be added to another agent as a connector, and that agent's actions can then call execute_agent to invoke a sub-agent. Useful for orchestration patterns where one high-level agent coordinates a fleet of specialized ones.

What happens if I disable Agent Center after wiring up this connector?
The connector stays enabled but list_agents returns empty and execute_agent fails with "no such agent" — because the catalog no longer exists. Re-enable Agent Center to restore full function.

What's next

Was this article helpful?