Agents may be allowed to
- Find permitted sources and structured context
- Draft a decision or action proposal
- Return provenance with the result
Agents · MCP · scoped access
Give agents useful knowledge—not blanket authority.
Connect agents to a Phasoric knowledge base through MCP with scoped access, explicit product contracts, and approval-first external action in Forge.
Let an agent retrieve permitted context or prepare a proposal without handing it an unbounded workspace token or the authority to make consequential changes silently.
Context and authority are different
An agent can know enough without being allowed to do everything.
Phasoric separates knowledge access, server authorization, and consequential action so each integration can expose only the capability it needs.
Connect an agent
Use documented MCP methods and developer contracts instead of browser secrets or an opaque all-access integration.
Grant a bounded scope
Authorize the permitted workspace and capability. Read access does not silently imply mutation authority.
Approve the effect in Forge
Preview supported consequential actions, confirm the exact effect, and retain an audit trail and recovery context.
The practical boundary
A governed path from context to effect
Agents do not automatically receive
- Every vault or every note
- Provider bearer tokens in the browser
- Approval to perform a consequential external write
MCP availability, scopes, and quotas vary by plan and capability maturity. Review the developer contract for the current supported surface.
Questions worth asking
Understand the tradeoff before you commit.
What can an MCP-connected agent access?
Only the methods, workspace mappings, and scopes authorized for that connection. A readable knowledge mapping does not automatically grant write authority.
Does MCP bypass Forge approval?
No. Where Phasoric promises approval-first action, an agent can prepare a proposal but the supported consequential effect remains behind the Forge approval boundary.
Can an agent see where its answer came from?
Supported retrieval contracts return source-linked context so the agent or reviewer can trace the result back to permitted knowledge instead of treating generated text as its own evidence.
Build on a narrower contract
Connect the agent. Bound the knowledge. Review the action.
Start with the MCP guide, then expose only the access your workflow actually requires.
Read the MCP guide