How to build Plain support dashboards in Metabase

Plain is an API-first B2B support platform — a GraphQL API, a typed SDK, and an MCP server are first-class, not afterthoughts. Metabase is where you turn that activity into shared, trustworthy dashboards. This guide covers two complementary paths: a lightweight MCP route for fast, AI-assisted questions, and a durable pipeline route that syncs Plain into a database so you can build dashboards anyone can read.

How do you connect Plain to Metabase?

Most teams combine these: use the MCP route for thread and tenant lookups, and the pipeline route for the dashboards people depend on.

What can you analyze from Plain data in Metabase?

• Thread volume — created vs. resolved by day and channel
• Time to first response — overall and by tenant
• SLA status — threads within and outside target
• Tenant health — open threads and load by company
• Backlog and aging — open work and how long it's been waiting
• Drivers — volume by label and custom thread field
• Channel mix — email, chat, and API-created threads

Which Plain dashboards should you build in Metabase?

How do you use the Plain and Metabase MCP servers together?

Pair the Plain MCP server with the Metabase MCP server for live, conversational analysis. The Plain MCP reads current threads, customers, and tenants; the Metabase MCP queries the models and dashboards you've already built.

Example workflows

• Show the open queue for a specific tenant and summarize by label.
• Trace a customer issue from a Plain thread to a GitHub PR and a Linear issue in one conversation.
• Cross-check tenant response trends against a Metabase model.

Be honest about the limits

• MCP is great for live lookups — not for scheduled or audited reporting.
• It does not create history; trend analysis still needs synced data.
• Write actions (reply, assign, change priority) ask for confirmation in your client.
• The Metabase MCP server is built in; an admin enables it under Admin → AI → MCP.

How do you set up the Plain and Metabase MCP servers?

{
  "mcpServers": {
    "plain": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.plain.com/mcp"]
    },
    "metabase": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://your-metabase.example.com/api/metabase-mcp"]
    }
  }
}

No API keys to manage — Plain's MCP uses OAuth, so the assistant inherits your Plain user's permissions. On first connection the server opens a browser window to authorize.

Can you generate a Plain dashboard with AI?

Yes — and this is the fastest way to a strong first draft. Use the prompt below with the Metabase MCP server and any assistant that can inspect your warehouse schema and create Metabase questions. It assumes Plain data is already synced into a database Metabase can read, treats MCP as exploratory, and tells the agent to skip metrics the schema can't support instead of faking them.

Create a polished Metabase dashboard for Plain support analytics using the
available Plain tables in this database.

Goal: Help support and customer success leaders understand volume, responsiveness,
SLA, tenant health, and contact drivers from Plain data.

First, inspect the schema and identify the available Plain tables. Do not assume
exact table names. Map the available raw tables into these analytical concepts
where possible: Threads, Timeline entries (messages/events), Customers, Tenants
(companies), Labels, Thread fields, and SLA status if present.

Important:
- Treat MCP data access as exploratory only.
- Build the dashboard from durable database/warehouse tables.
- Use medians (p50) and p90 for response times, never averages.
- Plain models companies as tenants — roll metrics up to the tenant where useful.
- Define "first response" as the first outbound message from a human or machine
  user, excluding internal notes.
- If timeline history is missing, do not calculate time-in-status. Use a caveat
  instead.
- Do not claim Metabase connects natively to Plain unless that is explicitly
  true in this environment.

Dashboard title: Plain Support Overview

Sections:
1. Executive summary (KPI cards): Threads created last 7 days; Resolved last 7
   days; Open backlog; Median time to first response; SLA status; Volume by
   channel.
2. Volume & backlog: Created vs resolved by day; Open by status; Backlog aging;
   Volume by channel.
3. Tenant health: Open threads by tenant; Tenants with rising volume; Response
   time by tenant; Top tenants by load.
4. SLA & response time: SLA status breakdown; First response p50/p90 by week;
   Reopened by week (only if history exists).
5. Drivers: Volume by label; Threads by custom field; Feature requests by tenant.

Filters: Tenant, Channel, Label, Assignee, Status, Date range.

Before finalizing, create or recommend reusable Metabase models:
modeled_plain_threads, modeled_plain_timeline_entries, modeled_plain_customers,
and modeled_plain_tenants.

Output: Build the dashboard if you have permission; otherwise provide the exact
questions, SQL, model definitions, and layout. Include caveats for any metric
that cannot be calculated from the available schema. Keep it practical, dense,
and executive-readable. Avoid vanity metrics.

How do you sync Plain data into a database or warehouse?

For dashboards that need history and reliability, land Plain data in a database first, then connect Metabase to that database.

Connector options

• Plain GraphQL API(raw) — the source of truth; query threads, timeline entries, customers, and tenants and upsert on a schedule. Plain advertises no restrictive rate limits.
• dlt(code) — wrap the GraphQL API in a Python pipeline for incremental loads and schema control.
• Typed SDK — Plain ships an open-source TypeScript SDK you can use to build a custom sync job.
• Webhooks — subscribe to thread events to keep your warehouse fresh in near real time.

Notes

• Land raw tables first, then build clean models on top.
• Sync timeline entries (not just threads) so you can compute response and resolution times.
• Bring tenants and thread fields into your model for per-account analysis.

How should you model Plain data in Metabase?

Core tables

Modeling advice

• Roll threads up to the tenant for B2B account-health dashboards.
• Normalize status (todo/snoozed/done) and channel so charts stay stable.
• Use actor_type on timeline entries to separate human, machine (AI), and customer messages for honest response metrics.
• Treat labels as a bridge table so a thread can carry many labels.
• Define "done" once and reuse it everywhere.

Which Plain metrics should you track in Metabase?

What SQL powers Plain dashboards in Metabase?

These assume the modeled tables above (PostgreSQL dialect). Adjust identifiers to match your warehouse.

SELECT
  date_trunc('day', t.created_at) AS day,
  COUNT(*)                                          AS created,
  COUNT(*) FILTER (WHERE t.status = 'done')         AS resolved
FROM threads t
WHERE t.created_at >= CURRENT_DATE - INTERVAL '30 days'
GROUP BY 1
ORDER BY 1;

SELECT
  tn.name            AS tenant,
  COUNT(*)           AS open_threads
FROM threads t
JOIN tenants tn ON tn.id = t.tenant_id
WHERE t.status <> 'done'
GROUP BY tn.name
ORDER BY open_threads DESC
LIMIT 25;

WITH first_outbound AS (
  SELECT
    e.thread_id,
    MIN(e.created_at) AS first_reply_at
  FROM timeline_entries e
  WHERE e.entry_type = 'message'
    AND e.direction = 'outbound'
  GROUP BY e.thread_id
)
SELECT
  date_trunc('week', t.created_at) AS week,
  percentile_cont(0.5) WITHIN GROUP (
    ORDER BY EXTRACT(EPOCH FROM (f.first_reply_at - t.created_at)) / 60.0
  ) AS median_first_reply_min
FROM threads t
JOIN first_outbound f ON f.thread_id = t.id
GROUP BY 1
ORDER BY 1;

SELECT
  l.label_type       AS label,
  COUNT(*)           AS threads
FROM threads t
JOIN thread_labels tl ON tl.thread_id = t.id
JOIN labels l ON l.id = tl.label_id
WHERE t.created_at >= CURRENT_DATE - INTERVAL '90 days'
GROUP BY l.label_type
ORDER BY threads DESC
LIMIT 20;

What are common mistakes when analyzing Plain in Metabase?

Related

Related analytics

Related integrations

FAQ