NotesSQLclMCPAI agents

SQLcl 26.3's MCP server adds RAG tools. Pin its restrict level

SQLcl 26.3 adds context_retrieve, annotation_generate and skills_sync for agents. Set the MCP restrict level yourself instead of trusting the default.

SQLcl 26.3 (September 2026) adds three AI tools to its MCP server for agents that work with Oracle schemas.

What's new

  • context_retrieve: runs a vector search that returns semantically relevant data (RAG). The new model command feeds it by vectorizing documents into database-managed tables.
  • annotation_generate: lets agents annotate tables and columns. A new annotation command can also annotate database objects automatically.
  • skills_sync: installs and updates AI skills.
  • Also in the changelog: 50% faster startup, PKCE wallet support, and 30+ Codescan rules for APEXlang (.apx) files.

Check the restrict level

The SQLcl 26.1.2 changelog says that, from that release, MCP starts unrestricted by default (previously level 4), while the 26.3 docs still say level 4 when no -R flag is given. Don't rely on either. Pin it: put "args": ["-R","4","-mcp"] in your MCP client config for every connection, and lower it only for a single task that needs it. Level 4 blocks host, scripting and configuration commands.

My take: new agent tools are a reason to tighten the defaults, not loosen them. Sources: SQLcl changelog and Configuring restrict levels for the SQLcl MCP server.

Get the Oracle + AI digest

One email when a new tested guide, template or APEX release note lands — English or Arabic. No spam; unsubscribe anytime.

← All notes