Data retention policyPAGE · CUSTOMERS MAY READ
Solutions · MCP knowledge gateway
Context your AI agents can be trusted with.
One governed MCP gateway feeds Claude, ChatGPT, coding agents, and your own AI infrastructure the same permission-aware knowledge.
C
ClaudeMCP CLIENT
ACTING FORalex@yourco · customers
tool callsearch("data retention policy") ◇ Identity⌘ Groups✓ Permissionschecked before retrieval
Security overviewPAGE · PUBLIC
Internal legal notesNOT PERMITTED · NOT RETRIEVED
The direct answer
How does GraphStaff govern MCP retrieval?
Each MCP request carries the identity and group context of the person or service asking. GraphStaff checks those permissions before retrieving knowledge, excludes restricted pages from the context, and returns permitted sources through one model-independent gateway that works with Claude, ChatGPT, coding agents, and custom AI infrastructure.
What platform teams do with GraphStaff
Stop pasting context into prompts.
Publish one gateway
Every AI agent connects to the same MCP server instead of a per-team scrape.
Enforce permissions in retrieval
Tool calls carry the reader's groups, so AI agents receive only permitted content.
Stay model-independent
Swap the model or the AI agent. The knowledge layer does not change.
Update once
Fix the page and every AI agent answers correctly on the next call.
Policy-checked retrieval
The access decision travels with the answer.
The policy engine that gates search and Ask AI gates retrieval, so an AI agent acting for Alex sees exactly what Alex sees.
Questions, answered
Frequently asked questions.
The short version of what teams usually want to know before they see GraphStaff in action.
Which AI agents can connect?
Anything that speaks Model Context Protocol: Claude, ChatGPT, Cursor, and custom agents.
Is this a separate product from the company brain?
No. This is the GraphStaff company brain seen from the AI platform side. The same governed knowledge graph that answers your people serves your approved AI agents through one permission-aware MCP gateway.
Can AI agents write through the gateway?
No. The MCP server is read-only by design.
How do you keep it model-independent?
The knowledge layer lives behind one governed MCP gateway, not inside any single model. You can swap Claude, ChatGPT, or an internal model without rebuilding the knowledge layer underneath.
See it with your knowledge
AI agents are only as good as their context.
Tell us which AI agents your teams run. We will show them answering from one governed source.