Not every integration needs custom code. If your agent just needs to query a database, search an index, or call an API that already has a spec, a connector gets you there with configuration only.
What’s in the catalog today:
- SQL databases: PostgreSQL, MySQL, MariaDB, Microsoft SQL Server, Oracle
- Solace Event Mesh: each operation becomes a publish or request-reply tool
- Amazon Bedrock knowledge bases
- Slack
- Custom: OpenAPI (any API with a spec) and Remote MCP (any MCP server)
- Other: Elasticsearch, OpenSearch, MongoDB, DynamoDB, Neo4j, Neptune
Here’s the Postgres connector from my demo setup, pointing at a read-only replica of a transportation management system:
kind: connector
name: northwind-tms-db
description: "Read-only replica of the Northwind TMS. Shipments, bookings, carrier assignments, milestones, and freight invoices."
spec:
type: sql
subtype: postgres
values:
hostname: tms-replica.northwind.example.com
port: 5432
database: tms_reporting
username: sam_readonly
password: ${NW_TMS_DB_PASSWORD}
A few tips from setting these up:
- Write the
descriptionfor the agent. The model uses it to decide when to reach for the connector, so say what data lives there, not just what the system is called. - Use a read-only user unless the agent really needs to write. It’s the easiest guardrail you’ll ever add.
- Keep secrets out of the file. The
${VAR}reference resolves at apply time, so the YAML can live in git.
Once a connector exists you can attach it to as many agents as you want. The list shows how many agents use each one, which helps when you’re about to change credentials. Workflows can call a connector directly too, using a tool node with no LLM in the loop.
Docs:
Download the Solace Agent Mesh desktop app and connect it to a database you already have.
Which data source would you connect first? And is there a connector you wish was in that list? Tell me in the replies.