IBM Knowledge Catalog is IBM’s governance catalog: catalogs and projects, seven governance-artifact types, a business glossary, data quality rules, and IBM Manta Data Lineage. It is authoritative for what IBM can observe and enforce on an IBM query path. An enterprise context layer, the category Atlan builds, is authoritative for what the business agreed, on every system and every agent vendor. The name forked rather than retired: the product became IBM watsonx.data intelligence on 2 May 2025 on IBM Cloud and SaaS, while Cloud Pak for Data 5.3 still requires the IBM Knowledge Catalog service for catalogs.
Agentic Data Intelligence arrived for SaaS in April 2026 and reached self-managed deployments on 17 June 2026, and IBM shipped an MCP server along the way. An agent can read IBM’s catalog through it today. What decides an architecture is how far IBM’s authority reaches once that agent acts on what it read, which is where a data catalog and a context layer start to diverge.
| Dimension | IBM Knowledge Catalog and IBM watsonx.data intelligence | Enterprise context layer |
|---|---|---|
| Authoritative for | What IBM observes and enforces on its own query path | What the business agreed, approved and owns |
| Observes | IBM-managed sources, plus 30-plus lineage-import sources | Systems inside and outside one vendor’s estate |
| Enforces for | Queries through IBM Data Virtualization or watsonx.data behind Presto | Named owners, wherever the asset lives |
| Agent interface | watsonx.data intelligence MCP server, remote and local | Hosted MCP server, APIs and SQL |
| Cost model | Resource units metered as records, tables, scans, subscriptions, pages | Platform subscription |
| Best fit | Context IBM produces about the estate IBM runs | One definition that resolves the same for every agent vendor |
What is IBM Knowledge Catalog?
It is the governance and cataloging service inside IBM’s data platform, supplying the containers, vocabulary and lineage everything else in that platform reads from.
Catalogs, categories and projects are its primary objects: Cloud Pak for Data 5.3’s overview states, twice, that catalogs and categories each require the IBM Knowledge Catalog service. The governance vocabulary is a fixed taxonomy of seven artifact types (categories, business terms, reference datasets, policies, governance rules, data classes and classifications), enumerated inside the definition of a billable record on IBM’s service plans doc, last updated 2026-08-21. On top sit the business glossary and Knowledge Accelerators, which IBM’s governance catalog page describes as ready-to-use vocabularies.
One boundary trips up readers first: IBM watsonx.governance is a different product that governs AI models, not data, and the two are not interchangeable in a procurement conversation.
Read against AI workloads rather than analyst workflows, the question changes shape. What IKC produces is strong catalog capability for AI workloads, and the gap that matters sits between what an agent reads and what a catalog stores. An agent does not browse a catalog page. It asks, gets an answer, and acts.
Is IBM Knowledge Catalog now called IBM watsonx.data intelligence?
Partly. The honest answer depends on which track you are on.
| Date | What changed | Track |
|---|---|---|
| 2024-07-03 | Cloud Pak for Data 5.0.0 releases; Watson Knowledge Catalog and InfoSphere Information Server migrate into IBM Knowledge Catalog | Both |
| 2025-05-02 | IKC renamed IBM watsonx.data intelligence; legacy plans stay valid, cannot be provisioned | SaaS and IBM Cloud only |
| 2026-04 | Agentic Data Intelligence becomes available | SaaS |
| 2026-06-17 | Agentic Data Intelligence extends to self-managed deployments | Self-managed |
IBM documents the rename on its legacy service plans page, last updated 2026-04-30, precisely enough to quote: “Prior to 2 May 2025, IBM watsonx.data intelligence was known as IBM Knowledge Catalog. If you provisioned IBM Knowledge Catalog and did not upgrade to a new plan after 2 May 2025, you have a legacy plan.” The 2024 migration and its 3 July 2024 release date sit on IBM’s WKC and InfoSphere migration page; the Agentic Data Intelligence dates come from IBM’s June 2026 announcement.
Now the counter-evidence, which almost nobody publishes: Cloud Pak for Data 5.3 states, present tense, that catalogs require the IBM Knowledge Catalog service. Our reading of it: the SaaS product name changed and the software component name did not, so a sentence that calls IKC “now watsonx.data intelligence,” full stop, is wrong on one of the two tracks. IBM’s docs are consistent per track; it is third-party listings and search indexes that trail, which is how a retired name still surfaces in a shortlist or an agent’s answer.
What IBM Knowledge Catalog does natively
A great deal, and the statuses worth carrying are the ones IBM publishes rather than the ones a reader infers from silence.
Import, then enrich, then assess quality: enrichment is LLM-powered, Data Product Hub handles data products and subscriptions, and version 2.3.0 added Open Data Contract Standard v3 contracts, per IBM’s what’s new in watsonx.data intelligence 2.3.x.
Lineage is where IBM’s native claim is strongest. IBM Manta Data Lineage is the engine underneath; IBM’s supported connectors for lineage import doc names more than 30 sources across cloud warehouses, relational databases, BI tools, ETL platforms and OpenLineage. IBM’s data lineage page claims historical lineage at the column level: IBM’s claim, not ours. Version 2.3.0 also added export of lineage to a third-party governance catalog. Good output, but an agent needs it queryable at runtime, across systems IBM never scanned, which is the AI-ready lineage test.
Three published statuses matter more than the feature list: Text-to-SQL is a “technology preview,” not supported for production; the v1/v2 data quality and v2 data profile endpoints were removed at 2.3.0; and on IBM Power, Cloud Pak for Data 5.2 lists Data Quality, Relationship Explorer and Semantic Enrichment as “not yet supported.”
A strong catalog. Whether it is agent-ready context is a separate test, as is whether data quality for AI agents holds when an agent writes a number rather than when a rule runs.
Does IBM Knowledge Catalog have an MCP server, and which agents can use it?
Yes. IBM documents a watsonx.data intelligence MCP server, remote and local, exposing 13 functional areas to named third-party clients.
In IBM’s own words, a client can “securely access, explore your data, and complete tasks across data governance and catalog, data quality, data lineage, and Data Product Hub through natural language.” Remote means a hosted endpoint on IBM Cloud or AWS; local runs on your machine; API credentials are required either way, per IBM’s MCP server doc. The 13 areas span connection management, data product management, data protection and governance, data quality, lineage, semantic search and query, text-to-SQL, metadata enrichment and import, project and container management, search and discovery, reporting, and workflow management. Named clients: IBM Bob, Claude Desktop, watsonx Orchestrate and GitHub Copilot in Visual Studio Code. IBM has open-sourced the implementation as IBM/data-intelligence-mcp-server under Apache-2.0.
IBM documents this server without publishing a GA or preview designation; we read the doc on 2026-09-16 and no sentence on it asserts a release stage, so treat it as documented, not guaranteed. Keep it distinct from Agentic Data Integration, whose own MCP server for the data agent carries IBM’s label on the integrating with the data agent doc: “This is a technology preview and is not yet supported for use in production environments.” That distinction matters when a procurement question is really a production-support question.
So an MCP-connected data catalog is not a differentiator anyone can claim any more. What separates implementations is whose estate the server can speak for, and whether the policy that governs an asset travels with the answer into the runtime an agent executes in. An IBM shop evaluating the Microsoft-side agent platform meets the same question.
Where IBM enforces its rules, and why its catalog reaches further than its authority
IBM answers where its rules are enforced precisely, and the answer is narrower than most readers assume.
Per IBM’s planning to protect data with rules doc, “Data protection rules are enforced in catalogs when all prevailing conditions for enforcement are met.” Then comes the sentence that sets the boundary:
“If you did not configure a deep enforcement solution, no assets are enforced in ungoverned catalogs or projects.”
Our reading of it: a rule that exists and a rule that stops a query are two different states, and IBM tells you which is which. The deep enforcement solutions IBM names are its own query paths, IBM Data Virtualization and IBM watsonx.data, and inside watsonx.data it narrows further. Per IBM’s connecting to IBM Knowledge Catalog doc, policies are definable only for the Presto (C++) and Presto (Java) engines, and row-level rules are supported across a published list of 11 connectors.
Put the two lists side by side: more than 30 lineage-import sources against 11 enforcement connectors, two engines with policy support and two deep enforcement solutions: the asymmetry is measurable, not asserted. IKC can describe far more of an estate than it can govern. For a human analyst, that gap is a caveat they will never notice; for an agent, it is the difference between retrieving a definition and being constrained by one.
IBM enforces where IBM owns the execution path, and documents where that path runs out. On a single-vendor estate the boundary may never bind; where agents reach data through engines IBM does not run, the policy must live where the engine cannot route around it. That is the same limit that stops role-based access control in an AI context platform, a governed metrics surface, another platform-native catalog and Google Cloud’s catalog at their own boundary. Estates spanning vendors need a multi-cloud context layer or cross-cloud governance above a native catalog.
What IBM’s meter counts, and why the measure is changing
IBM’s product page offers a free trial and names no price: Standard and Premium pricing means contacting IBM. The one public list price, with its provenance stated: on AWS Marketplace, a 12-month Essentials contract runs $30,840.00, plus $7.00 per unit of Premium-tier consumption.
The meter itself, per IBM’s service plans doc: a Record is one governance artifact or asset, a Table one lineage-scanned table, a Data Source Definition a system to scan, a subscription one access grant, a Page one document processed. Only the Capacity Unit-Hour measures time. Not one billable unit is an answer delivered.
IBM’s own legacy-plan guidance is explicit: delete unneeded assets and catalogs, because you are billed for the highest number that exist in a calendar month. Priced by how much it documents, a catalog has a standing incentive to document less, the same arithmetic behind how the split plays out on AWS and a neutral context layer over single-stack lock-in.
Coverage was the right measure while humans read catalogs. It changes when the reader is an agent that acts on what it retrieves.
| The shift | 2020 | 2023 | Now (2026) |
|---|---|---|---|
| Metadata supply | Manual | Automated | Autonomous |
| Metadata consumption | Siloed | Collaborative | Conversational |
| Flexibility and learning | Closed | Extensible | Self-improving |
IBM places cleanly in it: automated supply, conversational consumption via the MCP server, extensible learning via open APIs, ODCS v3 contracts and lineage export, but nothing self-improving.
An agentic catalog is measured by whether an agent found what it needed and, if not, why: never documented, retrieval needs work, or the model fell short. Four things carry that measurement: Context Agents creating context from evidence, a Context Lakehouse storing and vectorizing it, MCP serving it, and a learning loop routing usage back into creation. IBM documents semantic search and text-to-SQL at technology preview; vector, semantic, hybrid and graph traversal over one store is a different claim, which is why context engineering became its own discipline. An inventory and a traversable graph are not the same thing, nor is a semantic layer; together, why AI agents need an enterprise context layer beats a bigger catalog.
Real customers, real stories: Modern data catalog in action
53 % less engineering workload and 20 % higher data-user satisfaction
"Kiwi.com has transformed its data governance by consolidating thousands of data assets into 58 discoverable data products using Atlan. 'Atlan reduced our central engineering workload by 53 % and improved data user satisfaction by 20 %,' Kiwi.com shared. Atlan's intuitive interface streamlines access to essential information like ownership, contracts, and data quality issues, driving efficient governance across teams."
Data Team
Kiwi.com
🎧 Listen to podcast: How Kiwi.com Unified Its Stack with Atlan
One trusted home for every KPI and dashboard
"Contentsquare relies on Atlan to power its data governance and support Business Intelligence efforts. Otavio Leite Bastos, Global Data Governance Lead, explained, 'Atlan is the home for every KPI and dashboard, making data simple and trustworthy.' With Atlan's integration with Monte Carlo, Contentsquare has improved data quality communication across stakeholders, ensuring effective governance across their entire data estate."
Otavio Leite Bastos, Global Data Governance Lead
Contentsquare
🎧 Listen to podcast: Contentsquare's Data Renaissance with Atlan
How IBM Knowledge Catalog and an enterprise context layer divide the work
The division is additive, and each side supplies what the other cannot observe.
Assign authority per object, not per product. IBM owns schema on IBM-managed sources, lineage IBM Manta imports, Data Product Hub grants, and enforcement on an IBM query path, because enforcement happens where the query executes. The neutral layer owns the reconciled definition, the approval behind it, and context on systems IBM never observed: approval is a business event with a named owner. Skip that and tool-registration order sets the default precedence, since nothing flags two correct-in-their-own-system glossary entries as a conflict. The coexistence pattern for another platform-native surface follows the same sequence, the practical half of governing AI agents across multiple clouds.
Atlan is the Context Layer for AI: its homepage puts the mechanism plainly, more than 80 connectors pulling context into one living graph. The Atlan MCP server is hosted on every tenant, nothing to install, and every action it takes respects permissions already set in Atlan. Grounding an agent this way changes measurable output: agents grounded in governance metadata achieve 38% higher SQL accuracy than agents working from raw schema alone, per Atlan AI Labs. A context layer reference architecture shows where the two meet; how to implement one sets the sequence.
Some estates do not need the second layer yet: if your data sits in Db2 and watsonx.data behind Presto, every lineage source you need is on IBM’s import list, and no definition has to hold identically on an unobserved system, the bundled answer is defensible, until a second query engine, agent vendor or unscanned system changes it. Both vendors are credited by the same analyst: IBM is a Leader in the 2025 Gartner Magic Quadrant for Metadata Management Solutions, and Atlan is a Leader in the same report. IBM’s boundary is an architectural fact about execution paths, not a scorecard. That is the question platform-native context layers keep failing on estates with more than one vendor.
FAQs about IBM Knowledge Catalog
1. Did IBM Knowledge Catalog get renamed, and does the old name still apply anywhere?
Yes to both. IBM renamed it IBM watsonx.data intelligence on 2 May 2025, on the IBM Cloud and SaaS track. It is still the required component name in Cloud Pak for Data 5.3, where IBM states that catalogs and categories require the IBM Knowledge Catalog service.
2. Does IBM Knowledge Catalog have an MCP server, and which agents can use it?
Yes. IBM documents a watsonx.data intelligence MCP server, remote on IBM Cloud or AWS and local, exposing 13 functional areas. The clients IBM names are IBM Bob, Claude Desktop, watsonx Orchestrate and GitHub Copilot in Visual Studio Code. IBM publishes no GA or preview designation for it.
3. Where are IBM’s data protection rules enforced, and where are they not?
They are enforced in governed catalogs when all prevailing conditions for enforcement are met. IBM also states that without a deep enforcement solution configured, no assets are enforced in ungoverned catalogs or projects. The deep enforcement solutions IBM names are IBM Data Virtualization and IBM watsonx.data.
4. How much does IBM watsonx.data intelligence cost?
IBM publishes no list prices on ibm.com. The AWS Marketplace listing prices a 12-month Essentials contract at $30,840.00, with additional Premium-tier consumption at $7.00 per unit. Standard and Premium pricing means contacting IBM.
5. What is the difference between IBM Knowledge Catalog and IBM watsonx.governance?
Different products. IBM Knowledge Catalog governs data through catalogs, governance artifacts, quality and lineage. IBM watsonx.governance governs AI models through monitoring, factsheets and bias testing. Buying one does not give you the other.
Sources
Every URL below was read on 2026-09-16. Where IBM publishes a doc’s own last-updated date, that date is given.
- Service plans, IBM. Metering, updated 2026-08-21
- Legacy service plans, IBM. Rename quote and billing, updated 2026-04-30
- Planning to protect data with rules, IBM
- Connecting to IBM Knowledge Catalog, IBM Cloud
- Getting started with the MCP server, IBM
- Integrating with the data agent, IBM
- What’s new in 2.3.x, IBM
- Supported connectors for lineage import, IBM
- Cloud Pak for Data 5.3 overview, IBM
- WKC and InfoSphere migration, IBM Support. Released 3 July 2024
- Cloud Pak for Data 5.2, IBM. IBM Power gaps
- Agentic Data Intelligence, IBM. Dated 2026-06-17
- watsonx.data intelligence as a Service, AWS Marketplace
- Data lineage, IBM
- Governance catalog, IBM
- watsonx.data intelligence, IBM
- IBM/data-intelligence-mcp-server, GitHub. Apache-2.0
- Atlan MCP server, Atlan
- Atlan, a Leader in the 2025 Gartner Magic Quadrant for Metadata Management Solutions
