Skip to content
Sub-Agents vs MCP: Passive tools vs Active governance

Sub-Agents vs MCP: Passive tools vs Active governance

PT | EN

Have you heard of the Model Context Protocol (MCP)? It’s an open standard that allows AIs (like Claude) to use external tools. But being “just an MCP tool” is not enough for Vectora.

What are MCP Tools?

Imagine MCP is a Swiss Army knife. Each tool (like read_file or google_search) is a blade. The primary Agent (Claude) opens the blade it thinks it needs and uses it.

The problem? The primary Agent is not an expert in your infrastructure. It might try to read a giant file and crash the context, or ignore an important security rule.

The Sub-Agent Difference (Vectora)

Vectora is not just the knife; it is a Specialist Sub-Agent. When Claude calls it, it’s not just asking for a file, it’s delegating a context mission.

FeatureCommon MCP ToolVectora Sub-Agent
SecurityPrompt-dependent (fragile)Hard-coded Guardian (law)
EmbeddingsUsually noneIntegrated native pipeline
ValidationNoneHarness (Precision metrics)
NamespacesDirect disk accessIsolation via RBAC
DecisionPrimary Agent decides everythingContext Engine filters first

Protocols: MCP and ACP

Vectora speaks both languages:

  • MCP: To connect to any modern Tier 1 agent (Claude Code, Gemini CLI, etc.).
  • ACP (Agent Client Protocol): A proprietary ultra-low latency protocol for deep integration with your IDE (VS Code, Cursor).

Conclusion

Passive tools give functions. The Vectora Sub-Agent gives governance. It ensures your primary Agent receives the best possible context, in the safest and cheapest way, without you needing to configure every detail.


This is a supporting post for the Vectora project.