Platform
The data-context layer your AI agents have been missing
Strattum is a three-component middleware stack: an MCP connector registry that normalizes access to enterprise data sources, a policy enforcement engine that governs what each agent can see, and an audit event store that logs every context request for compliance.
MCP Connector Model
Connectors built for enterprise data contracts, not demo APIs
Most tool integrations for AI agents are thin API wrappers: they authenticate and pass queries through. Strattum MCP connectors do more. Each connector negotiates the schema of the source at connection time, so your agent always queries against a known, validated structure.
When a source schema changes, the connector surfaces the delta before your agent runs a query, not after it receives a malformed response. This is credential isolation at the connector level: the agent never holds the underlying credentials and never has access to the raw connection string.
Schema negotiation
Connector validates source schema at mount time and surfaces field-level changes before query execution.
Credential isolation
Service credentials are stored in the connector vault. The agent receives a scoped token, not direct database access.
Query parameterization
Agents submit parameterized context requests; the connector builds the source-specific query safely, preventing injection paths.
Incremental sync
For high-frequency context needs, connectors support watermark-based incremental reads rather than full re-scans.
Policy Enforcement Engine
Access rules defined once, enforced on every context request
The policy engine sits between the connector registry and your agents. Every context request is evaluated against the policy ruleset before the connector executes. No query reaches a data source without passing policy validation.
Policies are defined per agent identity and per connector scope. An agent can have a read-only view of Salesforce opportunity data, with a field-level exclusion on compensation records, and a separate allow rule for PostgreSQL analytics queries limited to a named schema. These rules are versioned and audited when changed.
Data classification tags from your source systems flow through the policy engine. If a record carries a sensitivity tag at the source, the engine can apply the corresponding allow or deny rule automatically without requiring the agent to know the classification logic.
Available controls
- Per-agent identity binding with OIDC or API key
- Per-connector scope declarations (connector, schema, table, field)
- Read-only enforcement at the connector level
- Data classification label pass-through and rule binding
- Policy version history with diff view
- Rule conflict detection on policy save
Example: access policy definition
policy "sales-agent-crm-read" {
agent_id = "sales-agent-v2"
connector = "salesforce-main"
allow {
objects = ["Opportunity", "Account"]
fields = "*"
mode = "read"
}
deny {
fields = ["Compensation__c"]
tag = "PII_HIGH"
}
}Audit Event Store
A complete record of what your agents accessed and when
Every context request generates a structured audit event. The store retains a tamper-evident sequence log that compliance teams can query, export, and deliver to auditors without custom extraction work.
Per-event fields
Agent identifier, connector ID, query scope, matched policy rule, response summary (not content), resolution status, and millisecond timestamp on every event.
Export formats
JSON event stream compatible with standard SIEM ingest (Splunk, Elastic, Datadog). CSV summary reports for non-technical compliance reviewers. Scheduled or on-demand.
SOC 2 use cases
Access review: answer "which agents read this data and when" in seconds. Incident investigation: replay event sequence for a specific agent and time window. Change evidence: policy version diff linked to audit events.
Audit event schema (JSON)
{
"event_id": "evt_01j4k9m7r3bx8f2c",
"ts": "2026-06-12T14:22:07.341Z",
"agent_id": "sales-agent-v2",
"connector_id": "salesforce-main",
"query_scope": { "object": "Opportunity", "fields": ["Name", "Stage"] },
"policy_rule": "sales-agent-crm-read/allow:Opportunity",
"resolution": "allowed",
"rows_returned": 12,
"duration_ms": 84
}Ready to give your agents governed context?
Follow the quickstart to connect your first data source in under four hours.