sqldw-authoring-cli

Telemetry — MANDATORY. Every api.fabric.microsoft.com call must carry x-ms-fabric-skill: sqldw-authoring-cli (az rest: --headers "x-ms-fabric-skill=sqldw-authoring-cli"), including every LRO poll, fabric_lro and retry. Snippets omit it — add it anyway.

Update Check — ONCE PER SESSION (mandatory) The first time this skill is used in a session, run the check-updates skill before proceeding.

• GitHub Copilot CLI / VS Code: invoke the check-updates skill.
• Claude Code / Cowork / Cursor / Windsurf / Codex: compare local vs remote package.json version.
• Skip if the check was already performed earlier in this session.

CRITICAL NOTES

1. To find the workspace details (including its ID) from workspace name: list all workspaces and, then, use JMESPath filtering
2. To find the item details (including its ID) from workspace ID, item type, and item name: list all items of that type in that workspace and, then, use JMESPath filtering

SQL Endpoint Authoring — CLI Skill

⚠️ SQL Execution Override: For SQL data-plane execution, this skill supersedes COMMON-CLI SQL/TDS guidance. Use MCP fabric-sqlendpoint-execute_query (see Tool Stack) unless explicitly using Legacy CLI Fallback.

Table of Contents

Tool Stack

IMPORTANT — MCP vs sqlcmd: This skill uses the fabric-sqlendpoint-execute_query MCP tool for all T-SQL execution. Do not use COMMON-CLI SQL/TDS/sqlcmd sections for query execution.

Agent preflight — verify before first operation:

1. Confirm the fabric-sqlendpoint-execute_query tool is available in your tool list. This tool is provided by the fabric-sqlendpoint MCP server, which is registered either by installing a Fabric skills plugin (the path for end users) or via this repo's .mcp.json — other MCP clients may register it through their own configuration.
2. If no matching tool is found, the user must register the Fabric SQL Endpoint MCP server. See mcp-setup/.
   • Global URL: https://api.fabric.microsoft.com/v1/mcp/dataPlane/sqlEndpoint
   • Item-scoped URL: https://api.fabric.microsoft.com/v1/mcp/dataPlane/workspaces/{workspaceId}/items/{itemId}/sqlEndpoint

MCP Tool Signature

fabric-sqlendpoint-execute_query(workspaceId, itemId, query)

Tool name may differ: execute_query is the logical operation. Depending on how the server is registered, the concrete tool name in your tool list may be prefixed (e.g. fabric-sqlendpoint-execute_query or sqlendpoint-global-execute_query). Invoke the concrete name shown in your tool list, always passing workspaceId, itemId, and query.

Returns: CSV resource (RFC 4180) with tabular results + metadata text.

Batch guidance: Multiple statements without GO are allowed in one call (e.g., CREATE TABLE ...; INSERT INTO ...). However, only the last result set is returned, and an error in any statement fails the entire batch. Prefer separate fabric-sqlendpoint-execute_query calls for independent DDL/DML operations — this gives clearer error messages and lets you verify each step succeeded before proceeding.

MCP Limits

These values are observed defaults, not a documented contract — the MCP service can change them. Treat them as guidance and confirm the current behavior from live 429 / timeout / truncation responses (or Microsoft Learn, if/when published) rather than relying on the exact numbers.

Authoring Scope by Item Type

Fabric DW DDL Constraints

These constraints are hard requirements — violating them produces errors:

MERGE and ALTER COLUMN are NOT hard errors. Per T-SQL surface area, MERGE is a generally available Warehouse feature and ALTER TABLE ... ALTER COLUMN is in preview. Prefer an explicit DELETE+INSERT when snapshot-conflict isolation matters, and CTAS + sp_rename for production-critical column-type changes — as a robustness choice, not because the syntax is blocked.

Correct CREATE TABLE pattern:

CREATE TABLE dbo.Orders (
    OrderID INT NOT NULL,
    CustomerName VARCHAR(100) NULL,
    Amount DECIMAL(19

Additional supported patterns:

• CREATE TABLE [dbo].[clone] AS CLONE OF [dbo].[source] — duplicate table structure + data
• COPY INTO — highest-throughput ingestion from external storage

Connection

Discover workspaceId and itemId

You need the workspace GUID and item GUID to call fabric-sqlendpoint-execute_query:

Execute a Query

No additional connection setup needed — authentication is handled transparently by the MCP protocol.

Verifying DDL/DML Results

For DDL (CREATE/ALTER/DROP), the tool returns success with metadata. Always verify:

Agentic Workflows

Schema Discovery Before Authoring

Before any write operation, discover the target schema:

Agentic Workflow

1. Discover → Run steps 1–4 to understand available tables/columns.
2. Sample → SELECT TOP 5 on relevant tables.
3. Formulate → Select pattern from SQLDW-AUTHORING-CORE.md (Table DDL through Common Authoring Patterns).
4. Execute → Call fabric-sqlendpoint-execute_query(workspaceId, itemId, query). For multi-batch operations (e.g., CREATE PROCEDURE with BEGIN/END), use a single batch without GO.
5. Verify → Query affected table (SELECT COUNT(*), SELECT TOP 5).
6. Optionally follow up → Run additional queries to confirm schema changes.

Gotchas, Rules, Troubleshooting

For full authoring gotchas: SQLDW-AUTHORING-CORE.md Authoring Gotchas and Troubleshooting. For CLI-specific issues: COMMON-CLI.md Gotchas & Troubleshooting (CLI-Specific).

MUST DO

• Verify workspace has capacity before creating warehouse — call GET /v1/workspaces/{id} and check capacityId.
• Verify fabric-sqlendpoint-execute_query MCP tool is available — check the tool list before the first operation. If unavailable, instruct the user to register the MCP server.
• Discover workspaceId and itemId first — resolve the target Warehouse via az rest; the tool takes GUIDs, not an FQDN or -d <DatabaseName>.
• az login first (for discovery) — the az rest workspace/warehouse lookups need an Azure CLI session. The fabric-sqlendpoint MCP server itself ships headerless and authenticates via your MCP client's native Fabric session, not the Azure CLI token; no signed-in session → auth failure on either path.
• SET NOCOUNT ON; in scripts — suppresses row-count messages that corrupt output.
• Send a single T-SQL batch per call — no GO separators and no -i file.sql; split multi-batch work (CREATE PROCEDURE, multi-step transactions) into separate fabric-sqlendpoint-execute_query calls.
• Label authoring queries with OPTION (LABEL = 'ETL_description').
• Use explicit CAST() in CTAS to control output types.
• Keep transactions short — long transactions increase conflict window.

AVOID

• GO separators — the MCP tool accepts only a single T-SQL batch. Combine related DDL in one statement or call fabric-sqlendpoint-execute_query multiple times.
• sqlcmd meta-commands (:setvar, :r, -i) — not available in MCP tool. Inline all SQL in the query parameter.
• Unbounded SELECT * — 10,000 row limit. Always use TOP N or WHERE to limit result sets.
• Singleton INSERT ... VALUES at scale — creates tiny Parquet files. Use INSERT...SELECT, CTAS, or COPY INTO.
• DROP TABLE IF EXISTS + CREATE TABLE to refresh — loses time-travel history. Use TRUNCATE TABLE + INSERT INTO.
• MERGE in production — GA, but table-level snapshot-conflict detection makes concurrent writers likely to fail. Prefer DELETE + INSERT when isolation matters.
• ALTER COLUMN in production — in preview; prefer CTAS + sp_rename for production-critical column-type changes (Schema Evolution).
• Variables in CTAS — not allowed. Wrap in dynamic SQL: EXEC sp_executesql N'CREATE TABLE ...'.
• DML on Lakehouse/Mirrored DB SQLEP — read-only for table data. Only views/funcs/procs can be authored.
• Concurrent UPDATE/DELETE on same table — snapshot isolation conflicts at table level. Serialize writes.
• Rapid-fire MCP calls — rate limit is 20 req/min. Consolidate multiple statements into one batch where possible.
• MARS — not supported. Remove MultipleActiveResultSets from connection strings.

PREFER

• CTAS over CREATE TABLE + INSERT — parallel, single-operation.
• INSERT ... SELECT over singleton INSERTs.
• COPY INTO for external file ingestion — highest throughput.
• DELETE + INSERT over MERGE for upserts in production.
• TRUNCATE TABLE over DELETE FROM without WHERE — faster, preserves history.
• Consolidating related DDL into a single fabric-sqlendpoint-execute_query call when no GO is required between statements.
• CTAS + sp_rename for large-scale transforms instead of UPDATE.
• fabric-sqlendpoint-execute_query MCP tool over sqlcmd for all T-SQL operations.
• SET NOCOUNT ON; prefix — reduces metadata noise in results.
• TOP N or WHERE clauses — stay within 10K row limit.

TROUBLESHOOTING