← Back to all guides

Databases & Spreadsheets — Airtable, Google Sheets & SQL

Langoedge Team5 min read

Giving an Agent Structured Data

To keep application data, customer leads, and operation logs structured without hosting a separate database, Langoedge agents can read and write Airtable and Google Sheets, or query a real relational database directly.

There are two routes, and they work quite differently:

Route Use for How it's set up
MCP Connector Airtable, Google Sheets, Notion, Coda, and other hosted data apps. Connect the account, add a connector node.
SQL toolkit Your own Postgres or SQL Server database. Enabled per deployment — see below.

1. Airtable & Google Sheets (via MCP Connector)

Airtable and Google Sheets are reached through an MCP Connector, not through built-in tools.

Earlier versions of this page listed individually named tools such as `airtable_list_bases` and `gsheets_read_rows`. Those were hand-written wrappers that no longer exist. The connector now discovers each app's tools live from the provider, which is why this page describes capabilities rather than fixed tool names — the authoritative list is whatever your connector shows once the account is linked.
1

Connect the account

On the **Connect** page, search for Airtable or Google Sheets and click **Connect Account**. You authorise directly with the provider.
2

Add an MCP Connector

On a Text Graph canvas, add an **MCP Connector** node and select the app. It loads that app's full live tool catalogue.
3

Scope and attach

The connector exposes every tool the app offers by default. For a database an agent writes to, consider turning off the destructive ones — a delete-records tool is rarely something you want an LLM reaching for unprompted. Then select the connector from your step's tool list.

What these apps let an agent do

The exact tool names come from the provider, but the shape of the capability is stable:

Airtable — list bases and tables, read and search records (including formula-based filters), create rows, update rows, upsert on a key, and delete records.

Google Sheets — list spreadsheets and worksheets, read a range or whole worksheet, append rows, update existing rows, and find rows matching a query.

**Appending vs. overwriting matters.** Both apps offer distinct "add" and "update" operations. If your agent is writing a log, expose only the append-style tool and withhold the update and delete ones — that removes a whole class of accidental data loss rather than relying on the prompt to prevent it.

2. Relational Databases (SQL Toolkit)

For a real Postgres or SQL Server database, Langoedge has a SQL toolkit that lets an agent inspect the schema and run queries.

**The SQL toolkit is disabled by default.** It only loads when the deployment sets `ENABLE_SQL_TOOLS=true`, and its tools are registered lazily on first use. If you don't see SQL tools in your tool picker, that flag is off for your instance — contact your administrator or Langoedge support rather than assuming the feature is missing.

When enabled, the toolkit registers a set of standard LangChain SQL tools for listing tables, inspecting schemas, checking a query, and executing it. Connection credentials are URL-encoded and handled server-side; the LLM never sees them.

Best Practices

Withhold destructive tools

Use the connector's tool scoping to turn off delete and bulk-update operations unless the agent genuinely needs them. Prompt instructions are a weaker guarantee than simply not exposing the tool.

Give the agent IDs, not guesses

Most write operations need a base, table, or spreadsheet ID. Let the agent list them first rather than hardcoding IDs into a prompt that will break when someone renames something.

Keep writes off the voice loop

MCP connectors are Text Graph only. A voice agent should call the Text Graph asynchronously so a slow spreadsheet write doesn't leave the caller in silence.

Summarize before returning

A raw record dump burns context. Put a summarization step after a large read so only the relevant fields re-enter the conversation state.

Frequently Asked Questions

Why don't I see airtable_list_bases in my tool list?
Because it no longer exists. Airtable is reached through an MCP Connector, whose tool names are supplied by the provider at run time. Add the connector, pick Airtable, and the node will show you the current tool list.
How do I search records efficiently in Airtable?
Airtable's MCP tools support its formula syntax, so a filter such as `{Email} = 'user@example.com'` works as expected. The connector will show the exact parameter name once loaded.
Will an agent overwrite my spreadsheet?
Only if you give it a tool that can. Append and update are separate operations — scope the connector to expose just the append tool if the sheet is an append-only log.
Can I use these on a Voice Graph?
Not directly — MCP connectors are Text Graph only. Put the connector on a Text Graph and mount that graph as a tool on your voice node.
Is the SQL toolkit available on my account?
Only if your deployment has `ENABLE_SQL_TOOLS` turned on. It is off by default.
LT

Langoedge Team

The Langoedge engineering team builds AI agent infrastructure that empowers businesses to deploy reliable, observable AI staff. Follow Langoedge Team on LinkedIn for product updates and architectural deep dives.