Skip to content

ROADMAP.md

Latest commit

 

History

History
56 lines (45 loc) · 2.5 KB

ROADMAP.md

File metadata and controls

56 lines (45 loc) · 2.5 KB
Raw

Roadmap

The roadmap is capability- and quality-gated. Dates are intentionally omitted until the project has a stable maintainer cadence.

Phase 1 — Safe read-only foundation

  • Typed Python package and CLI entry point
  • stdio and Streamable HTTP transports
  • Databricks unified authentication
  • Health, current identity, catalog, and warehouse tools
  • Deny-by-default policy foundation
  • Unit tests, CI, container, and security documentation
  • Capability-pack architecture with a policy-enforcing tool registry
  • Capability introspection (server_info, list_enabled_capabilities, get_policy_status)
  • Unity Catalog metadata exploration (schemas, tables, views, columns, functions, volumes)
  • Mock-SDK contract tests for every read tool
  • Read-only SQL with sqlglot AST validation, bounded results, and cancellation
  • Read-only jobs, pipeline, and cluster inspection
  • Read-only governance (grants and object permissions)
  • Read-only workspace browse/export, MLflow, Model Serving, Vector Search, Genie
  • Read-only cost and audit reads over system tables
  • Consistent cursor pagination for large listings
  • Opt-in live-workspace integration tests

Phase 2 — Controlled operations

  • Mutation safety contract: risk classes, dry runs, approval tokens, audit log
  • controlled-write access mode with per-tool allowlist (writes off by default)
  • First gated mutation pack: jobs run / cancel / delete
  • Pipelines, compute, and warehouse lifecycle operations
  • Workspace and Unity Catalog Volume file operations
  • Idempotency-key store and async task support for long-running operations

Phase 3 — Enterprise remote server

  • MCP OAuth 2.1 protected-resource implementation
  • M2M, U2M, on-behalf-of-user, and workload identity federation
  • Multi-user and multi-workspace isolation
  • OpenTelemetry, audit sinks, quotas, and distributed rate limiting
  • Databricks Apps, Docker, Kubernetes, and Helm deployments

Phase 4 — Platform coverage

  • Unity Catalog lineage, grants, and governance
  • MLflow experiments and model registry
  • Model serving, Genie, AI Search, and agent tooling
  • Account-level identity, workspace administration, and cost visibility
  • Public extension SDK for third-party capability packs

Phase 5 — Version 1.0

  • Stable tool and response contracts
  • Client compatibility and upgrade guarantees
  • Security review, SBOM, provenance, and signed releases
  • PyPI, container registry, MCP Registry, and Marketplace publication