write_db_table¶
Write a table into a saved live-database connection.
The source is a file or an open tab; the target is a schema and table on the server. Use it to land a cleaned or converted result somewhere other tools can reach.
Write tool. Removed from the roster entirely when the server runs
with --mcp-read-only.
Parameters¶
| Name | Type | Meaning |
|---|---|---|
connection |
string (required) | Saved connection name or id |
schema |
string (required) | Target schema (public, dbo, or the database name on MySQL) |
table |
string (required) | Target table name |
mode |
string | create (default, error if the table exists), append, replace (DROP + CREATE) |
path |
string | Source file, any supported format. Cloud URLs included |
open_tab |
string | Source open tab (in-GUI assistant only): @active, a tab name, or #2 |
catalog |
string | Top namespace level on Snowflake, Databricks and BigQuery. An error elsewhere |
Give exactly one of path or open_tab.
Result¶
The write is gated¶
Refused unless the connection's Allow writes switch is on. That
switch is off by default and is per connection, so a read-only reporting
connection cannot be written to by accident — check it first with
list_db_connections.
replace drops the existing table before recreating it. There is no
undo on a live server: prefer create when the table should not already
exist, and let the error tell you when it does.
Types¶
Columns are created from the source's Arrow types through the same
dialect mapping export_schema uses, so the DDL
matches what that tool would have rendered for the engine. Identifiers
are quoted in the CREATE and the INSERT identically, so a column with
a capital letter or a space survives the round trip.
See also¶
copy_db_table— server to server, no file in the middle.sync_sql— get the SQL for review instead of writing.- Database Connections.