Connect over MCP

Use Rowset's MCP server when a trusted agent needs to discover Rowset tools and work with private datasets without browser automation.

If you are deciding whether your workflow belongs in Rowset, start with What is an agent-managed dataset?. If you are choosing between MCP and plain HTTP, read When should an AI agent use MCP instead of REST?. If you are deciding between delegated OAuth and a provisioned bearer key, use the MCP OAuth vs API keys decision guide.

1. Get the MCP URL

https://rowset.lvtd.dev/mcp/

The Rowset dashboard setup prompt includes this URL and your agent API key.

If you use a self-hosted instance, specify the public URL you expose for it, including the /mcp/ path. For example:

https://rowset.example.com/mcp/

Create the API key on the same instance you connect to.

2. Store the API key

Store the key in a private environment variable such as ROWSET_API_KEY, or in your client's secret store. Do not paste the raw key into public prompts, screenshots, issue trackers, or repositories.

MCP requests should send:

Authorization: Bearer YOUR_ROWSET_API_KEY

3. Add the server to the agent client

For Codex/OpenClaw-compatible clients, run:

codex mcp add rowset --url https://rowset.lvtd.dev/mcp/ --bearer-token-env-var ROWSET_API_KEY

The command stores the environment variable name. Make sure the agent runtime can read the full ROWSET_API_KEY value before the client starts.

If your client only supports custom headers, set Authorization to Bearer YOUR_ROWSET_API_KEY. Private REST requests use the same bearer authentication format.

4. Verify the connection

For a new connection, ask the agent to call:

get_user_info

get_user_info verifies the authenticated account, completes onboarding, and starts the trial. Repeat it only when diagnosing connection or credential problems. Do not load capabilities or list datasets merely because a session started. Use live MCP tool schemas for the operation at hand. Call get_rowset_capabilities only for an unfamiliar feature or troubleshooting, then request specific topics; opt into use cases or full mode only when needed.

5. Inspect before changing rows

If the user supplied a dataset key or URL, call get_dataset directly. When the relevant dataset is unknown, use a bounded search before creating a duplicate:

search_datasets {"query": "relevant name or purpose", "limit": 3}

Select a result, then call get_dataset before row operations. That response includes headers, index column, dataset instructions, schema metadata, project context, and relationship summaries. If you are designing those rules for a new workflow, use the guide to structuring dataset instructions for AI agents.

6. Create a small first dataset

When the agent is setting up a new workflow, have it read the relevant use case first, then create a dataset with a clear description, instructions, and stable index_column.

Useful starting points:

7. Use the right permission level

Agent API key permissions apply to MCP tools:

Read keys can inspect account details, projects, datasets, rows, and exports.
Read + write keys can also create and update datasets, rows, projects, relationships, schema, and public preview settings.
Admin keys can also create new agent API keys through REST or MCP.

Permissions are account-wide, not scoped to one dataset or operation. Read + write also permits destructive tools such as deleting rows, dropping columns, removing relationships, and archiving datasets. Filter the tools available to a narrowly scoped agent in its runtime or application, and ask before destructive actions such as deleting rows, archiving datasets, or clearing public preview passwords.

Direct database MCP servers

Direct database MCP servers are better when an agent must query an existing Postgres, MySQL, SQLite, or warehouse system and the operator is ready to manage database credentials, permissions, query cost, and schema safety.

Rowset is narrower. Use it when the agent needs its own private backend for structured workflow data for task boards, CRMs, feedback queues, catalogs, QA trackers, content pipelines, or similar structured row workflows. If you are comparing Rowset with open-source database workspaces, read Baserow alternatives for AI-agent-managed datasets. If your team is choosing between spreadsheet-style backends, see NocoDB alternatives for AI-agent-managed datasets.

Read MCP Database: Direct Access vs Agent-Managed Data for the longer decision guide, including permission boundaries and stable row identity.