Decision FrameworkDecision layer

MCP Census: Buy vs Skip (MCP Server Vetting for Client Agent Stacks)

IF your agency ships agent stacks that install third-party MCP servers into client environments and you need a repeatable PASS/REVIEW/BLOCK/UNKNOWN gate, THEN start on the free tier (1,000 credits each UTC month, 10 anonymous lookups/day per IP) and move to builder at $19/month for 2,000 credits plus REST, MCP, stamp, and watches once audits run across more than a handful of client configs. IF you need 10,000 credits each UTC month for multi-client CI preflight and change feeds, the Team plan at $49/month is the break point; anything beyond server vetting (malware scanning, pen testing, compliance certification) sits outside what Census returns.

By InnovaAI ResearchPublished

Decision Frame

MCP Census: Buy vs Skip (MCP Server Vetting for Client Agent Stacks)

“IF your agency ships agent stacks that install third-party MCP servers into client environments and you need a repeatable PASS/REVIEW/BLOCK/UNKNOWN gate, THEN start on the free tier (1,000 credits each UTC month, 10 anonymous lookups/day per IP) and move to builder at $19/month for 2,000 credits plus REST, MCP, stamp, and watches once audits run across more than a handful of client configs. IF you need 10,000 credits each UTC month for multi-client CI preflight and change feeds, the Team plan at $49/month is the break point; anything beyond server vetting (malware scanning, pen testing, compliance certification) sits outside what Census returns.”

When is it the right choice?
  • Your delivery team runs .mcp.json audits across several client repos and needs one verdict per entry via the /v1/audit endpoint rather than manual GitHub and npm checks
  • Client agent stacks span Claude Code, Codex, Cursor, VS Code, or GitHub Actions and you want the Census rule embedded in CLAUDE.md, AGENTS.md, or .cursor/rules so servers get stamped before install
  • You resell a one-time gate as a productized offer, such as the $1,800 MCP Census SMB Audit at roughly 16h setup, and need stamped evidence artifacts to hand the client
  • Ongoing registry drift matters to a retainer, so the builder tier's watches and change feed (cursor-based, since-last-check) justify $19/month over ad hoc re-checks
  • Your agency wants metered spend it can forecast: purchased builder credits never expire, so unused capacity carries across quiet months
When should you skip it?
  • The engagement requires malware scanning, penetration testing, or formal compliance attestation, which Census explicitly does not provide and cannot be positioned as covering
  • Your agents only call first-party or internally built MCP servers, so a public-evidence gate returns little beyond PASS and the free 1,000 credits go unused
  • Volume stays under the free tier's 1,000 credits each UTC month and 10 anonymous lookups/day per IP, making any paid plan pure overhead
  • No client workflow touches MCP servers at all, so the REST and MCP APIs have nothing to audit and the $49 Team tier buys capacity you will not consume
  • The client insists on a single vendor security platform rather than a point gate, and a four-verdict lookup tool would fragment their existing review process
ai-infrastructure