What’s included
What’s included
Classification and access
Sensitivity labelling, masking and fine-grained access control on AWS.
Built with
- Amazon SageMaker
- AWS Lake Formation
- AWS Glue Data Quality
- Amazon Macie
- Amazon Bedrock AgentCore
Why it matters
Why it matters.
AI amplifies whatever your data already is.
Poor quality and unclear sensitivity become customer-visible problems fast.
Teams slow down when nobody can say whether a dataset is safe to use. Clear classification removes that hesitation.
Agents now reach data through MCP tools. Lineage, access control and a record of every call are what your auditors and enterprise customers ask for first.
What’s included · how it runs
Assess. Implement. Connect. Govern.
Step 01
Assess
Review quality, ownership, sensitivity and current controls.
What’s included
Classification and access
Sensitivity labelling, masking and fine-grained access control on AWS.
Step 02
Implement
Stand up the catalogue, lineage, quality tests and access controls.
What’s included
Catalogue and lineage
Discoverable datasets with owners, definitions and end-to-end lineage.
Quality framework
Automated quality tests, thresholds and alerting on the datasets that matter.
Step 03
Connect
Expose governed datasets to your AI agents as MCP tools, with sign-in and policy at the gateway.
What’s included
Governed MCP access
Curated datasets exposed to AI agents as MCP tools, with each person’s own permissions, policy at the gateway and every call audited.
AI readiness controls
Rules for what data may reach a model, with the evidence to prove they hold.
Step 04
Govern
Agree ownership, a review cadence and the policy for AI use of data.
What you walk away with
- Data catalogue with owners and lineage
- Automated data quality checks
- Classification and access controls
- Governed MCP tools for your AI agents
- AI data usage policy and evidence
Governed access for AI · MCP
AI agents get the data. Your rules still apply.
Teams want AI assistants, IDEs and their own agents to query company data through MCP, the Model Context Protocol. We expose governed datasets as MCP tools, so an agent sees only what the person asking is allowed to see.
- 01 · Who asks
A person
Signs in once with your identity provider.
- 02 · MCP client
AI assistant or agent
An IDE, a chat assistant or your own agent.
- 03 · Control point
MCP gateway
Amazon Bedrock AgentCore Gateway checks the sign-in and applies policy on every call.
- 04 · MCP tools
Read-only tools
One per approved dataset or question, over curated data.
- 05 · Governed data
Your data on AWS
AWS Lake Formation applies the person’s row and column grants.
On the record
Every call logged in AWS CloudTrail on behalf of the person, every policy decision in Amazon CloudWatch.
Tools over curated data
Catalogued datasets exposed as MCP tools through Amazon Bedrock AgentCore Gateway: read-only by default, described well enough for an agent to pick the right one.
The person’s permissions, not the agent’s
OAuth 2.1 sign-in, as the MCP specification sets out, and trusted identity propagation, so Lake Formation applies each person’s grants. The model never sees the token.
Policy at the gateway
Cedar rules in Policy in Amazon Bedrock AgentCore decide which agent can call which tool, with which parameters, outside the agent’s code.
Every call on the record
Approved MCP servers listed in AWS Agent Registry, IAM guardrails for AWS-managed MCP servers, and an audit trail your auditors can read.
01 / 04
Case studies
Customers we’ve delivered this for.
Case study[Client]
[The promise in 2 lines].
[How the client got the outcome: what Assembly built and where it runs. One sentence.]
Talk to an engineer[Client logo][First Last][Role]
Case study[Client]
[The promise in 2 lines].
[How the client got the outcome: what Assembly built and where it runs. One sentence.]
Talk to an engineer[Client logo][First Last][Role]