Integrating Tiger Cloud with My AI Agent via Tiger CLI and MCP
For Global Hack Week: Hacktoberfest I completed the Tiger challenge: Integrate Tiger Cloud with your AI Agent with the Tiger CLI and MCP.
What I built
Not an app — an agent-ready database setup:
- Installed Tiger CLI
v0.26.0on Windows (irm https://cli.tigerdata.com/install.ps1 | iex) - Authed with
tiger auth login(OAuth, projectNew Project) - Created a free Tiger Cloud service:
- Name:
ghw-hacktoberfest-agent-db - Type: TIMESCALEDB, Region:
us-east-1, Status:READY - Verified with
tiger service listandtiger db query -c "SELECT version();"→ PostgreSQL 18.6
- Name:
- Wired Tiger MCP to my agent (OpenCode) in
opencode.json:
"tiger": { "command": ["tiger", "mcp", "start"], "type": "local", "enabled": true }
- Restarted the agent so the
tigerMCP server loads (service_list,service_get,db_query,db_schema,search_docs,view_skill, + hypertable skills)
Why MCP + CLI
- CLI = explicit, scriptable control (
tiger service create/list,tiger db query/psql) - MCP = same operations as typed tools for the agent, plus docs search and best-practice skills (schema design, hypertable setup, migration planning)
- Both share one login, so anything I do in terminal the agent can do too
How I verified
-
tiger mcp listshows 30 tools/prompts -
tiger db schemaconnects to the new service - Dashboard shows the service READY: https://console.cloud.tigerdata.com/dashboard/services/pphbkrfp6g
- Submitted to GHW: Hacktoberfest challenge with dashboard + CLI screenshots
Takeaway
One binary gives you both interfaces. If you use OpenCode/Cursor/Claude Code and your client isn't in tiger mcp install, manual config with tiger mcp start works fine. For shared/prod DBs, consider tiger config set read_only prod.
Built for GHW Hacktoberfest 2026. Service: ghw-hacktoberfest-agent-db on Tiger Cloud.
Top comments (2)
Nice to see the CLI and MCP route written up end to end. One thing I'd be curious about: does the agent connect with a role that can only read, or can it write to the database too? That choice decides how much a bad prompt can cost you.
Great question, Nic — that distinction is huge when handing an LLM direct database access.
By default, Tiger MCP connects with the primary administrative role (tsdbadmin), meaning out of the box the agent does have full write and DDL permissions (INSERT, UPDATE, DELETE, DROP). The tool description for db_query even flags this with a destructive execution warning.
To prevent a bad prompt or runaway agent from wrecking schemas or mutating records, Tiger CLI & Cloud offer practical guardrails
In addition, Tiger MCP caps output rows with mcp_max_rows (default 100) to keep context tokens from blowing up on unindexed wide scans. For shared or production setups, setting read_only=prod or locking the connection down to an unprivileged role is definitely the way to go!