mcp-data-platform-v0.22.1

What's New

Fix: Knowledge Pipeline apply Now Supports Column-Level Targets

The apply_knowledge tool's apply action previously ignored the target field on update_description changes, silently writing all descriptions to the dataset level. Column targets like column:location_type_id were treated identically to empty targets.

Before (v0.22.0): apply with target: "column:email" updated the dataset description instead of the column description. No error was returned — the wrong entity was silently modified.

After (v0.22.1): The target field is parsed for the column: prefix and routed correctly:

Target value	Routed to	Example
"" (empty)	UpdateDescription (dataset-level)	Update the dataset's overall description
"column:email"	UpdateColumnDescription (column-level)	Update the email column's description
"column:address.zip_code"	UpdateColumnDescription (column-level)	Nested field paths work too

Mixed batches are fully supported — a single apply call can update dataset descriptions, column descriptions, tags, and glossary terms in one operation.

Enum Constraints in Tool Schemas

Both capture_insight and apply_knowledge now expose explicit JSON schemas with enum arrays for all categorical fields. Previously, LLM clients had to guess valid values through trial and error because the MCP Go SDK's auto-generated schemas from struct tags don't support enum constraints.

Fields with enum constraints:

Tool	Field	Valid values
capture_insight	category	correction, business_context, data_quality, usage_guidance, relationship, enhancement
capture_insight	confidence	high, medium, low
capture_insight	source	user, agent_discovery, enrichment_gap
Both tools	action_type / change_type	update_description, add_tag, add_glossary_term, flag_quality_issue, add_documentation
apply_knowledge	action	bulk_review, review, synthesize, apply, approve, reject

MCP clients that render tool input schemas (e.g., Claude Desktop, Claude Code) will now show dropdown/autocomplete options for these fields.

Improved Tool Descriptions

• apply_knowledge description now documents the column:<fieldPath> target format with examples
• flag_quality_issue behavior is documented: adds a quality_issue:<detail> tag to the entity
• capture_insight description clarifies the source field values (agent_discovery for self-discovered insights, enrichment_gap for flagging metadata gaps)

Dependency: mcp-datahub v0.7.2

Upgrades from v0.7.1 to v0.7.2, which includes:

• Null aspect crash fix: getAspect() no longer panics when DataHub returns HTTP 200 with a null/empty aspect value (entities where editable properties were never written)
• Column description writes: New Client.UpdateColumnDescription() method for read-modify-write on the editableSchemaMetadata REST API aspect

Changelog

Bug Fixes

• e064386 fix: knowledge pipeline apply — column targeting, enum schemas, mcp-datahub v0.7.2 (#117)

Upgrading

This is a backwards-compatible patch release. No configuration changes required.

Behavioral change: apply_knowledge with target: "column:..." now correctly writes to column-level descriptions instead of dataset-level. If you had workarounds for the previous behavior (e.g., calling DataHub REST API directly for column descriptions), those workarounds can be removed.

Installation

Homebrew (macOS)

brew install txn2/tap/mcp-data-platform

Claude Code CLI

claude mcp add mcp-data-platform -- mcp-data-platform

Docker

docker pull ghcr.io/txn2/mcp-data-platform:v0.22.1

Verification

All release artifacts are signed with Cosign. Verify with:

cosign verify-blob --bundle mcp-data-platform_0.22.1_linux_amd64.tar.gz.sigstore.json \
  mcp-data-platform_0.22.1_linux_amd64.tar.gz