Control MCP Server Access Across Developer Workstations
The Model Context Protocol lets AI agents call tools that read files, run commands, and reach local services. Cup’n’String can sit in the path of MCP activity to apply tool allowlists, restrict filesystem and shell access, shield secrets, and audit tool calls.
Why this matters
MCP turns developer workstations into integration hubs for AI agents. Each MCP server is a new capability surface that is easy to add and hard to inventory.
MCP servers can expose filesystem, shell, and network capabilities to agents
New servers are added per project, often without central review
Tool calls can reach local databases, APIs, and Docker-exposed services
Unmanaged MCP servers are difficult to inventory and audit
Sensitive output can leave the workstation through tool responses
The Cup’n’String approach
Cup’n’String treats MCP as a first-class governance surface. Where MCP activity is routed through supported paths, it can audit, allowlist, restrict, and shield.
Active Proxy & Shielding sits in the path of MCP tool calls to audit and, by policy, block disallowed actions
Tool allowlists and parameter boundaries constrain what agents may invoke
Filesystem and shell access can be restricted by policy
Sensitive output can be redacted in transit where content inspection is enabled
A governed MCP server registry supports administrative approval workflows
Review governed access requests, approve a narrower scope, and manage time-limited access leases
How it works
- Step 1AI agent / MCP client
- Step 2Cup’n’String MCP proxy
- Step 3Tool allowlist & policy check
- Step 4Approved tool / local service
- Step 5Audit & policy evidence
Checklist
- Can you inventory the MCP servers in use?
- Can you allowlist which MCP tools agents may call?
- Can you restrict filesystem and shell access?
- Can you audit MCP tool calls?
- Can you require approval for new MCP servers?
- Can you self-host the control plane?
Frequently asked questions
Does Cup’n’String support MCP?
Yes. MCP is a key governance surface. Where MCP activity is routed through supported paths, Cup’n’String can apply tool allowlists, restrict filesystem and shell access, shield secrets, and produce audit logs.
Can it block specific MCP tools?
With Active Proxy & Shielding in the path of MCP traffic, policy can block disallowed actions or destinations in real time. Teams typically start in observe mode and move high-confidence policies to enforce.
Does this replace the MCP servers themselves?
No. Cup’n’String governs the interaction between agents and MCP servers; it does not replace the tools. Developers keep their workflows while security gains visibility and control.
Can MCP activity be audited for compliance?
Yes. Governed MCP activity, tool usage, and policy decisions can be recorded to help security and platform teams understand how agents use tools.
Bring MCP under policy, approval, and audit
Govern how agents use MCP tools without blocking the workflows your developers rely on.