Planned AI connector
Give Claude controlled access to BidAgent tools through one secure MCP surface.
The planned BidAgent Claude connector will use remote MCP tools for authorized opportunity, pursuit, analysis, and compliance workflows.
Use the same governed tool layer
ChatGPT and Claude should connect to one canonical BidAgent MCP service rather than separate business-logic implementations.
- Shared tool schemas
- Consistent tenant authorization
- Consistent audit records
Prefer remote HTTP MCP
A public HTTPS streamable HTTP endpoint is the planned transport for the hosted connector, with tokens issued through the approved authorization flow.
- Streamable HTTP
- Bearer-token validation
- Server-side pagination and limits
Design for prompt-injection resistance
Solicitation content is untrusted input. Tools must keep document text separate from system instructions and constrain links, outputs, and write actions.
- Untrusted-content boundaries
- Structured output schemas
- Approval for material actions
Frequently asked questions
Is the Claude connector live?
Not yet. BidAgent is preparing a shared remote MCP service and must complete authentication, safety tests, rate limits, and connector review before release.
Will Claude use a separate BidAgent database?
No. The connector should use the same authorized BidAgent service layer and tenant controls rather than receiving direct database access.