Skip to main content
Glama

ssis-export-toolkit

An MCP server that lets an AI assistant (Claude) create, deploy, schedule and run SSIS packages that move SQL Server tables to Azure Blob Storage as CSV. Governed by design: one installation per SQL Server VM, one MCP per person, every action signed and policy-checked.

You give orders in plain language ("export dbo.Orders to Blob Storage every day at 03:00"). The assistant turns them into a declarative spec, generates the .dtsx packages and the project, and a small signed-request executor on the VM deploys them to SSISDB, runs them, verifies the result and creates the SQL Agent job. Nobody runs scripts by hand except the one-time installation by the administrator.

Claude + ssis-builder MCP (your PC) --signed request--> Blob Storage (ssis_control/) <--pull every minute-- executor (VM) --> SSISDB / SQL Agent

Why

  • Repetitive packages, generated. Export packages differ only in table, columns, types and file name. A spec describes the interface; the generator renders the package deterministically (same input, same output, diff-friendly).

  • Safe by construction. No secrets in packages or specs; the storage key lives in the VM's DPAPI store. Source queries must be SELECT-only. Requests are Ed25519-signed per person and checked against a policy that only an administrator can change.

  • Auditable. Every action records who did it (executor log, request status, job description).

Related MCP server: mcp-mssql

Who does what

Role

What

Guide

Administrator (once per VM)

install the executor, store the storage key, authorise colleagues, own the security rules

docs/ADMIN_GUIDE.md

User (once per PC)

clone, run scripts/ssis_builder_setup_pc.ps1, send your public key to the administrator

docs/USER_GUIDE.md

Everybody, on each release

git pull; VM-side changes need the administrator

docs/UPGRADING.md

Tools (MCP)

  • Specs and packages: ssis_spec_create, ssis_interface_spec_create, ssis_validate, ssis_generate_export, ssis_generate_bundle.

  • VM operations: ssis_executor_status, ssis_deploy, ssis_configure, ssis_run, ssis_verify, ssis_cleanup.

  • Jobs: ssis_schedule, ssis_job_enable, ssis_job_status, ssis_job_delete.

  • Administration and identity: ssis_executor_installer, ssis_invite_colleague, ssis_my_public_key.

Safety model (short)

Allowed operations, databases, object-name prefixes and containers are defined in a policy that lives on the VM. A request is rejected unless its signature verifies, it has not expired (max 60 min), it was never seen before (replay protection) and it passes the policy. Details: docs/executor.md.

Status

Early-stage. The spec, generators, signing and executor logic are covered by unit tests (simulated channel and SSISDB). Generated .dtsx files cannot be validated without SSDT/dtexec: run the first deployment against a test server. Imports (Blob Storage -> table) are specified but the import generator is not built yet; see package generators.

Develop

python -m venv .venv; .\.venv\Scripts\pip install -r requirements.txt
.\.venv\Scripts\python -m unittest discover -s tests

Docs: package generators, interface spec v2, executor.

License

MIT, see LICENSE.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • F
      license
      Not graded
      quality
      D
      maintenance
      Enables AI assistants to interact with SQL Server Data Warehouses using natural language for automatic schema discovery and report generation. It ensures security by restricting operations to read-only SELECT queries through both code validation and database permissions.
      -
    • A
      license
      A
      quality
      D
      maintenance
      Enables natural language to SQL queries on MSSQL databases via Claude, with safe SELECT-only execution and schema discovery.
      3
      ISC