MCP 2026 Compatibility & Migration Auditor

What does this Actor do?

MCP 2026 Compatibility & Migration Auditor checks public, unauthenticated MCP HTTP endpoints against the released Model Context Protocol specification 2026-07-28. It performs bounded read-only live discovery, evaluates a versioned ruleset, and returns one explainable report per canonical endpoint. Each report includes a compatibility status, a normalized 0–100 score, confidence and coverage, stable check IDs, official references, evidence, warnings, and prioritized migration actions.

The Actor only sends safe protocol discovery and capability-list requests. It never calls tools, reads resources, sends prompts, uses cookies or credentials, or follows a redirect without validating the destination again. Private, loopback, link-local, metadata, multicast, reserved, and unspecified destinations are rejected before any request.

Why use it?

Use this Actor when you need a repeatable migration inventory across several remote MCP servers. The output is structured for API consumers and AI workflows: stable serverKey values support deduplication, checkIds support remediation tracking, and official references make each result auditable. Unknown evidence stays unknown and lowers coverage/confidence instead of being presented as a definite compatibility failure.

How to use it

1. Add one to ten public MCP endpoint URLs in mcpServers.
2. Optionally add positional public GitHub repositories and enable includeRepositoryAnalysis for secondary static evidence.
3. Choose a timeout between 3 and 60 seconds.
4. Run the Actor and inspect the Dataset output. Start with compatibility.status, compatibility.score, dataQuality, then follow migrationActions and their linked check IDs.

Example input:

{
  "mcpServers": ["https://mcp.apify.com/?tools=docs"],
  "githubRepositories": [],
  "timeoutSecs": 15,
  "includeRepositoryAnalysis": false
}

Input

mcpServers accepts one to ten endpoint strings. Valid public HTTP(S) URLs are audited; malformed, unsupported, credential-bearing, or private entries produce isolated reports without an outbound request. Credentials, sensitive headers, unknown fields, and unbounded values are rejected. githubRepositories is optional and positionally associated with the server list. Repository inspection is unauthenticated, bounded, opt-in, and never overrides live protocol evidence.

Output

The Dataset contains one report per canonical endpoint. The main fields are serverKey, server, compatibility, capabilities, checks, migrationActions, warnings, and dataQuality.

Simplified output:

{
  "serverKey": "https://example.com/mcp",
  "compatibility": { "status": "partial", "score": 75, "targetSpecification": "2026-07-28" },
  "checks": [{ "id": "discovery.server-discover.required", "status": "pass" }],
  "migrationActions": [],
  "dataQuality": { "coverage": "complete", "unknownChecks": [] }
}

Download the Dataset as JSON, CSV, Excel, or HTML from Apify. JSON preserves the complete nested evidence and provenance contract; CSV and Excel are convenient for inventories and remediation tracking, while HTML is useful for sharing a readable snapshot.

Cost

This Actor does not set or change pricing automatically. Apify run cost depends on allocated compute, runtime, request volume, and optional repository analysis. The Actor itself caps endpoint count, requests, response bytes, redirects, retries, repository files, and concurrency.

Tips and limitations

• Use public endpoints that do not require authentication.
• Treat unknown and partial as data-quality signals, not automatic failures.
• Keep repository analysis disabled unless static evidence is useful.
• The audit is intentionally read-only and does not prove that tools are safe to invoke.

FAQ

Does it call MCP tools? No. Tool calls, resource reads, prompt retrieval, and other side-effecting methods are outside the audit boundary.

Why is a reachable server unknown? The server may not expose the bounded discovery shape, may return malformed data, or may exceed a safety limit. Inspect warnings and dataQuality.

Can it audit private network endpoints? No. This is a deliberate SSRF and privacy boundary.

API, support, and disclaimer

Use the Apify API to start a run and retrieve its default Dataset. The input and output schemas in .actor/ are the machine-readable contract. Report issues with a reproducible public endpoint, sanitized input, Actor run ID, and relevant warning/check IDs; never include credentials or tokens.

This tool provides bounded compatibility evidence, not certification, security approval, or legal advice. Always review the official MCP 2026-07-28 specification, server discovery, Streamable HTTP transport, and changelog.

Development

npm test
npm run build
npm run lint
apify validate-schema
npm run benchmark:fixtures

The fixture benchmark disables network access and reports runtime, memory, requests, retries, response bytes, endpoint count, and output count. Publishing and pricing remain manual operations.