# n8n — Last9 MCP

> Give n8n's AI Agent node access to your production telemetry via Last9's MCP server — query exceptions, traces, logs, and metrics from inside any workflow.

Source: https://last9.io/docs/integrations/n8n-mcp/

n8n's AI Agent node accepts **MCP Client Tools** as attachable tool nodes. Point one at Last9's MCP server and the agent gets the full Last9 toolset — exceptions, logs, traces, PromQL queries, alerts, change events — usable inside any workflow.

The classic shape: an alert webhook fires → an AI Agent node investigates with Last9 tools → the result posts to Slack. No human in the loop until the agent runs out of confidence.

:::note
This guide covers using Last9 as a data source **from** n8n workflows. If you want to instrument n8n itself and ship its execution traces **to** Last9, see [n8n + OpenTelemetry](/docs/integrations/n8n-otel/).
:::

## Prerequisites

- n8n 1.22+ (when MCP Client Tool support was added)
- A Last9 org slug and MCP token — from [Query Tokens](https://app.last9.io/control-plane/query-tokens) (Token Type: **Client** → Client Type: **MCP**)
- An LLM provider credential configured in n8n (Anthropic, OpenAI, etc.)

## Setup

1.  **Enable MCP tools in n8n**

    Set this environment variable on your n8n instance and restart:

    ```bash
    N8N_COMMUNITY_PACKAGES_ALLOW_TOOL_USAGE=true
    ```

    Without it, the MCP Client Tool node won't appear as an attachable tool in the AI Agent node.

    If you're running Docker Compose, add it to your `environment` block:

    ```yaml
    environment:
      N8N_COMMUNITY_PACKAGES_ALLOW_TOOL_USAGE: "true"
    ```

2.  **Get your Last9 org slug and create an MCP token**

    - **Org slug** — the path segment right after `app.last9.io/` in your browser (e.g. `https://app.last9.io/acme-corp/...` → `acme-corp`)
    - **MCP token** — from [Query Tokens](https://app.last9.io/control-plane/query-tokens) (admin required):
      1. Click **New Token**
      2. Set **Token Type** → `Client`
      3. Set **Client Type** → `MCP`
      4. Give it a name (e.g. `n8n alert triage`)
      5. Click **Create Token** and copy the value

3.  **Add a Bearer Auth credential in n8n**

    In n8n settings → **Credentials** → **New Credential** → **HTTP Bearer Auth**:

    | Field     | Value                           |
    | --------- | ------------------------------- |
    | **Token** | The MCP token you created above |

    Name it something recognizable, e.g. `Last9 MCP Token`.

4.  **Build a workflow with AI Agent + Last9 MCP**

    In the n8n editor:

    - Create or open a workflow
    - Add an **AI Agent** node
    - Attach an LLM to the agent (e.g. **Anthropic Chat Model** or **OpenAI Chat Model**)
    - Click the **Tools** input socket on the AI Agent node and add **MCP Client Tool**
    - Configure the MCP Client Tool:

      | Field                | Value                                                      |
      | -------------------- | ---------------------------------------------------------- |
      | **Transport**        | `HTTP Streamable` (preferred) or `SSE`                     |
      | **Endpoint**         | `https://app.last9.io/api/v4/organizations/<org_slug>/mcp` |
      | **Authentication**   | `Bearer Auth`                                              |
      | **Credential**       | Select the Last9 credential you created above              |
      | **Tools to Include** | `All` (or scope to specific tool names)                    |

5.  **Test it**

    Add a **Manual Trigger** node, connect it to the AI Agent, and click **Test workflow**. Set the agent's prompt to something that requires Last9 data:

    ```
    What exceptions has the payment-service thrown in the last 30 minutes?
    Group by exception type and include the most recent stack trace for each.
    ```

    The agent will call `get_exceptions`, receive the data from Last9, and write the summary into the workflow output. From there you can fan out to Slack, Linear, PagerDuty, or any other node.

## Workflow patterns

**Alert triage bot**

A webhook from your alerting system triggers the workflow. The AI Agent uses Last9 tools to fetch recent exceptions, check for recent deploys via change events, and look up trace error rates. It posts a structured triage summary to Slack with suspected root causes. Time-to-context drops from minutes to seconds.

```
Incoming alert: payment-service latency p99 > 2s

Investigate this alert using Last9. Check:
1. Recent exceptions in payment-service
2. Any change events (deploys, config changes) in the last 2 hours
3. Downstream service error rates

Summarize your findings and rate your confidence (low/medium/high)
that you've identified the root cause.
```

**Daily ops digest**

A cron trigger runs at 9 AM. The agent asks "which services degraded in the last 24 hours?" and sends the summary to the on-call Slack channel before standup.

**Deploy postmortem helper**

A GitHub webhook fires after a deploy. The workflow waits 15 minutes, then the agent compares pre/post error rates and latency using PromQL tools. It posts a verdict: thumbs-up or "investigate this" with specifics.

**On-call escalation filter**

An alert fires → agent checks if the anomaly is a known flap pattern using historical trace data → if it looks real, pages the on-call engineer; if it's a known pattern, acknowledges and closes the alert automatically.

## Available Last9 MCP tools

For the full list of tools and their parameters, see the [Last9 MCP](/docs/integrations/mcp/#available-tools) reference.

The agent picks the right tools automatically based on your prompt. You can constrain which tools are available by listing specific names in **Tools to Include** on the MCP Client Tool node.

## Complete working example

An importable n8n workflow JSON and Docker Compose setup are available in the [opentelemetry-examples](https://github.com/last9/opentelemetry-examples) repository under `n8n-mcp/`. The workflow covers the alert triage pattern end-to-end.

## Troubleshooting

**MCP Client Tool not selectable on the AI Agent node.**
You missed `N8N_COMMUNITY_PACKAGES_ALLOW_TOOL_USAGE=true` or didn't restart n8n after setting it.

**401 from Last9.**
The Bearer token must be a **Client → MCP** token from [Query Tokens](https://app.last9.io/control-plane/query-tokens). API keys from the API Access page, Refresh Tokens, and session tokens won't work on the MCP endpoint.

**Wrong org slug.**
The slug is the path segment immediately after `app.last9.io/` — not your team name, display name, or email address. Check your browser URL bar.

**Agent calls no tools.**
Prompt the agent more specifically. Open-ended prompts like "how is my system doing?" may not trigger tool use. Prompts that name a specific service, time window, or signal type work better.

**HTTP Streamable transport fails; SSE works.**
Some reverse proxies buffer responses and break SSE-over-HTTP streaming. Try `SSE` transport as a fallback, or configure your proxy to pass through `Transfer-Encoding: chunked`.

Please get in touch with us on [Discord](https://discord.com/invite/Q3p2EEucx9) or [Email](mailto:support@last9.io) if you have any questions.