Use Porter’s PostgreSQL connector for Claude to ask about the tables you choose through a read-only role, then compare them with Google Ads spend.
By Porter Metrics team · Last updated October 19, 2026
ClaudePorter connectedPorter / PostgreSQL / Trailgear Demo✓ 4 rows| Day | Rows |
|---|---|
| Oct 7 | 1,012 |
| Oct 8 | 1,047 |
| Oct 9 | 1,096 |
| Oct 10 | 1,284 |
One table per questionPorter’s PostgreSQL connector for Claude takes 3 steps, and you don’t install an MCP server anywhere. Create a read-only role and enter its credentials in Porter. Add Porter’s MCP URL to Claude as a custom connector, then ask about any table that role reads. MCP (Model Context Protocol) is the open standard Claude uses to connect to tools.
Run the SQL under “What the connector needs from PostgreSQL” to create a read-only role. Then sign up free at Porter, pick PostgreSQL as the source and enter that role’s credentials. Finally, pick Claude as the destination.
In Claude, open Customize, then Connectors, and choose Add custom connector. Then name it Porter Metrics and paste the URL of Porter’s MCP server:
https://mcp.portermetrics.com/mcpSign in to Porter when Claude asks, then paste:
PromptUsing my PostgreSQL table public.orders, count rows by day for the last 14 days using the created_at column.
Each use case comes with a prompt to copy, the columns it needs and who it’s for. Column names are examples, so swap in your own table and columns, and name the date column so Claude filters on the right one.
The daily count lands in your chat instead of a ticket for whoever has database access.
PromptFrom my PostgreSQL table public.orders, count rows and sum amount by day and status for the last 30 days, using created_at as the date. Which status changed most?For: RevOps and ops leads · founders · growth marketers
So you can decide which source gets next week’s budget without waiting on someone’s saved query.
PromptFrom my PostgreSQL table crm.leads, count leads by source for last week and last month using created_at, plus leads where status is won. Rank sources by won leads.For: growth marketers · sales ops · agencies
So a drop shows up the day you ask, not in the weekly reconciliation.
PromptFrom my PostgreSQL table public.orders, show daily row count and sum of amount for the last 60 days using created_at. Flag days more than 30% away from the trailing 7-day average.For: founders · product managers · ops
As a result, you get order totals by plan without a new view or a request to the backend team.
PromptFrom my PostgreSQL tables public.orders and public.customers, get last month's order count and amount by customer_id, then each customer's plan. Match them on customer_id and total orders by plan.For: RevOps · analysts · founders
The weekly budget call can then rest on the signup count the product team trusts, set against what paid search cost.
PromptFor the last 8 weeks, show Google Ads Cost by week next to the weekly count of rows in my PostgreSQL table public.users using created_at. Work out cost per signup and say whether higher-spend weeks brought more signups.For: growth and performance marketers · agencies
The PostgreSQL project publishes no official MCP, so Porter brings the tables your role can read into Claude, next to your ad, CRM and analytics data. Need to write SQL or change rows instead? Use a provider MCP such as Supabase’s MCP server or Neon’s MCP server, which run any SQL you allow.
Porter hosts the MCP server, so there’s nothing to install or patch. Create a read-only role, enter its credentials in Porter and add one MCP URL to Claude.
Claude reads every table your role can SELECT and tells you what changed, while your rows stay exactly as you set them. Your grants decide which tables Claude sees.
Ask about orders or signups next to Google Ads, Meta Ads, GA4, HubSpot and other Porter connectors, for cost per signup from the table your product team trusts.
Porter reaches any PostgreSQL that accepts its connection, including Supabase, Neon, Amazon RDS and Cloud SQL. It signs in with database credentials, not a provider login.
The PostgreSQL connector comes with every Porter plan, including Free, so your first table costs nothing to try.
You can start free, since Porter’s PostgreSQL connector for Claude comes with every plan. Use the calculator for your number of accounts, or see Porter pricing for every plan.
$33.33/mo$6.67 per account
An account is one ad account, Stripe account, store, page or GA4 property. The Free plan has unlimited accounts with 30 days of history.
The PostgreSQL connector gives Claude no fixed field list. Instead, each table your role can read shows up as its own account, named schema.table. Its columns then become the fields Claude filters, groups and adds up. Claude reads one table per question and matches results across tables in chat.
| What Porter reads | Example | Note |
|---|---|---|
| Tables | public.orders, analytics.signups | One account each |
| Columns | created_at, status, amount | Only columns the role can read |
| System columns | ctid | Never listed |
| What | PostgreSQL connector |
|---|---|
| Fields | No fixed list; each column the role can read is a field |
| Actions | Reads only, so your database stays exactly as you set it |
| Reads | One table per question, named schema.table, from every schema the role can read |
| History | 30 days on the Free plan, 15 in AI chats |
| Access | Read-only role with CONNECT, USAGE and SELECT |
| Not read | System columns such as ctid, and tables the role can't SELECT |
| Databases | Any PostgreSQL that accepts outside connections, including Supabase, Neon, Amazon RDS and Cloud SQL |
Yes. Porter reads through a role that holds only SELECT, so your rows and tables stay exactly as you set them, and PostgreSQL itself enforces that. If you can, point the connection at a read replica to keep long reporting questions off the primary.
Porter needs CONNECT on the database, USAGE on each schema and SELECT on each table (PostgreSQL's GRANT reference). For tables created after setup, the role that creates them runs ALTER DEFAULT PRIVILEGES (PostgreSQL's ALTER DEFAULT PRIVILEGES docs).
CREATE ROLE porter_ro LOGIN PASSWORD 'choose-a-strong-password'; GRANT CONNECT ON DATABASE mydb TO porter_ro; GRANT USAGE ON SCHEMA public TO porter_ro; GRANT SELECT ON ALL TABLES IN SCHEMA public TO porter_ro; ALTER DEFAULT PRIVILEGES IN SCHEMA public GRANT SELECT ON TABLES TO porter_ro;PostgreSQL only accepts remote connections once listen_addresses goes beyond localhost, and pg_hba.conf must allow the connection, ideally with scram-sha-256 passwords (PostgreSQL's pg_hba.conf rules). In addition, the strictest SSL mode, verify-full, checks both the server certificate and its host name (libpq's SSL modes).
A replica takes reporting load off the primary, especially for long questions. On a hot standby, though, PostgreSQL cancels a query that blocks replay once max_standby_streaming_delay passes, 30 seconds by default (PostgreSQL's standby server settings).
The questions fall into two groups. Three are about fit: hosted databases, whether MCP works with databases at all, and tables across several schemas. Then three are fixes: numbers that don't match your SQL client, tables with no columns, and long questions that fail.
Yes. Porter's PostgreSQL connector for Claude reaches a hosted database as long as it accepts outside connections. Porter signs in with database credentials, not a provider login, so Claude can read Supabase, Neon, Amazon RDS and Cloud SQL databases alike. Neon and Prisma Postgres also appear in Porter as separate integrations for actions.
Yes. In short, an MCP server such as Porter's sits between Claude and the database and decides what Claude can run. Porter's hosted server, for example, reads tables through a role you grant. However, provider servers from Supabase or Neon can run SQL, including writes, unless you switch on their read-only setting.
Yes. Porter lists tables from every schema your role can read under one connection, each named schema.table, and then Claude asks about whichever table you name. Grant USAGE only on the schemas Claude should see, because those are the only tables that show up.
Claude's numbers through Porter usually differ from your SQL client because the two read a different table, filter, date column or time zone. So check those four first. A read replica can also trail the primary, so a count taken there may miss the newest rows.
Grant SELECT on the table and USAGE on its schema, then ask Claude about the table again. PostgreSQL hides a table's columns from roles without a privilege on it, so an empty column list means the Porter role needs that grant.
Ask for a shorter date range. A long question can stop inside PostgreSQL before Porter returns any rows. The statement_timeout setting cancels statements past its limit (PostgreSQL's statement_timeout setting), and a hot standby can also cancel queries. Finally, the max_connections setting, typically 100, can refuse a new connection (PostgreSQL's connection settings).
The MCP URL behind the PostgreSQL connector also gives Claude every other source you connect in Porter. So your app tables can sit beside warehouse, CRM or analytics data in one chat. Two database siblings and three marketing sources pair especially well with them, or browse all Claude connectors:
Setting up the PostgreSQL connector for Claude takes one database role and one URL. Then ask which day dipped, which source converts and what a signup costs. The free plan already includes the MCP, so your first table costs nothing to try.
Written by the Porter Metrics team, who build Porter's Claude connectors.
Free plan · 14-day trial · No card needed
Growth marketers: cost per signup from the users table instead of an export.
RevOps and ops leads: daily counts by status and source, straight from the database.
Founders: a same-day read on unusual days in orders or signups.
→
→
→
→