Transport Parity Testing

MCP Rigor can execute the same acceptance suite against named stdio and Streamable HTTP targets, then compare normalized behavior test by test.

Plain-language parity for QA teams

Create calculator-parity.mcpr:

MCP Test 1

Suite: "Calculator behaves the same everywhere"

Compare target "Local": node server.js
Compare target "QA environment": https://qa.example.com/mcp

Test: "Adding 20 and 22 gives 42"
  Id: adding-numbers

  Call tool "add" with:
    a: 20
    b: 22

  Expect "structuredContent.sum" equals 42

Test: "The service status is ready"
  Id: service-status

  Read resource "service://status"
  Expect "contents[0].text" contains "ready"

Run it exactly like a YAML parity suite:

mcprigor parity calculator-parity.mcpr

The Compare target lines are usually configured once by an engineer. QA authors only add ordinary Test, Call tool, Read resource, and Expect statements. Each test automatically runs against every named target.

A target value beginning with http:// or https:// is Streamable HTTP. Any other value is parsed as a stdio command.

YAML alternative

Automation engineers may also use the structured YAML representation:

version: 1
name: Calculator parity

target: &default
  transport: stdio
  command: node
  args: [server.js]

targets:
  stdio:
    transport: stdio
    command: node
    args: [server.js]
  deployed:
    transport: streamable-http
    url: https://staging.example.com/mcp

tests:
  - id: add
    name: Addition is consistent
    steps:
      - tool:
          name: add
          arguments: { a: 20, b: 22 }
        assert:
          json:
            path: $.structuredContent.sum
            equals: 42

Run the matrix:

mcprigor parity parity.yaml
mcprigor parity parity.yaml --markdown --out parity.md

The first named target is the baseline. Every target runs independently with a fresh MCP session. MCP Rigor compares:

• test status;
• step status and MCP method;
• normalized response structures;
• captured/exported outputs.

It excludes durations, elapsed times, timestamps, session IDs, and _meta transport details. Application response fields are not ignored.

A divergent test includes a deterministic path-level semantic diff. The command exits 1 when any target differs.

Scope

Parity means equivalent observable behavior under the selected tests. It does not prove byte-level transport equivalence, identical concurrency behavior, network reliability, authentication parity, or production infrastructure correctness.