Agent-native data backend
Dataset backends for AI agents
Rowset gives trusted agents hosted MCP tools and a REST API for structured rows. They create datasets, keep indexes stable, update records, export snapshots, and share read-only previews when humans need to inspect the result.
Private by default. Bearer-key authenticated. Public previews are optional, read-only, and separate from MCP or REST access.
MCP URL
https://app.rowset.com/mcp/
You are a data operator.
Use Rowset MCP tools to manage rows.
Prefer stable_index for updates.
Do not guess IDs.
- Auth
- Bearer key
- Index
- stable_index
- Preview
- off
dataset
customer_events
| Key | Plan | Status | Next action |
|---|---|---|---|
| A182-0001 | pro | active | Check renewal |
| A182-0002 | free | trial | Send setup guide |
| A182-0003 | team | blocked | Confirm owner |
- MCP tools
- Create, read, update, delete rows
- REST API
- Predictable endpoints and bearer auth
- Stable index
- Business keys or generated rowset_id
- Private default
- Public previews only when enabled
- Exports
- CSV, JSONL, XLSX, SQLite, Parquet
Setup
Three steps to agent-ready data
The human handoff stays small. The agent owns the row work through authenticated tools.
-
01
Copy the setup prompt
Give your agent the MCP URL, REST base URL, dataset rules, and context.
ROWSET_API_KEY
stored privately
-
02
Authorize with a bearer key
Keys stay scoped to your account. The agent verifies access before creating data.
Authorization: Bearer -
03
Let the agent manage rows
It creates datasets, updates rows by stable keys, exports snapshots, and keeps previews explicit.
create_row update_row export_dataset
Operational layer
From file to API-backed rows
Rowset does not need to own source connectors. Your agent can read local files, sheets, databases, or exports with its own tools, then write the structured working set to Rowset.
customers.csv
The source can stay wherever it belongs.
Agent instruction
Maintain customer_events.
Add new customers.
Update plan changes.
Export weekly.
The agent translates context into stable row operations.
Rowset dataset
- Rows
- 5,231
- Index
- stable_index
- Access
- MCP + REST
Rowset returns consistent, queryable rows.
Why Rowset
Agents need state, not another fixed app
A dataset is enough structure for an agent to keep working without making a human operate a spreadsheet or build a backend first.
- Shape changes by job
- The agent creates headers, instructions, metadata, and schema details that fit the workflow.
- Indexes stay explicit
- Use a business key when one exists. Use rowset_id when the source has no stable identifier.
- Humans keep control
- The dashboard is for setup, current state, exports, settings, and public preview review.
- Private path stays private
- MCP and REST stay authenticated. Public previews are read-only browser sharing, not an auth shortcut.
Decision
Why not a spreadsheet or custom backend?
Spreadsheets are easy to inspect but awkward for agents to mutate safely. Custom backends are durable but slow to build for a new workflow. Rowset sits in the middle.
| Capability | Spreadsheet | Custom backend | Rowset |
|---|---|---|---|
| Agent-native MCP and REST | Manual | Weeks of work | Production-ready |
| Stable row handles | Easy to break | Custom to build | Built in |
| Private authenticated writes | Limited or manual | Complex to implement | Bearer-key access |
| Exports for handoff | Manual | Custom pipeline | CSV, JSONL, XLSX, SQLite, Parquet |
| Human review preview | Links and permissions | Custom to implement | Optional and read-only |
Use cases
Workflow-shaped datasets
Start with one row shape the agent can maintain without guessing. Expand when the workflow proves useful.
task_id
Agent task board
Shared task state across agent sessions, owners, blockers, and completion evidence.
feedback_id
Feedback triage
Deduplicate signals, preserve source links, assign themes, and export accepted rows.
Personal CRM
Let an agent keep follow-ups, relationship state, notes, and stale-contact reviews current.
Start with one dataset
Give your next agent workflow a real backend
Create an account, connect a trusted agent, and let it manage structured rows through MCP or REST.
codex mcp add rowset
ready--url https://app.rowset.com/mcp/
--bearer-token-env-var ROWSET_API_KEY
- First tool
- get_user_info
- First dataset
- create_dataset
- First handoff
- export_dataset