MCP Rigor logo MCP Rigor

Background: MCP testing landscape

Non-normative research. This page explains product positioning; it is not needed to use MCP Rigor.

Executive summary

MCP testing tools currently fall into five categories: interactive inspectors, scriptable probes, official protocol conformance, SDK/framework-local tests, and semantic or security evaluators. The strongest opening for MCP Rigor is a transport-neutral, deterministic, black-box application test runner: “Playwright for MCP.”

MCP Rigor should complement the Inspector and official Conformance project rather than suggest it replaces either.

Landscape

ToolPrimary roleWhat it does wellGap relative to MCP Rigor
MCP InspectorOfficial interactive debuggerExplore connections, capabilities, tools, resources, prompts, notifications, and logsPrimarily manual exploration rather than a scenario/assertion/reporting framework
MCP Inspector CLIScriptable probeProgrammatic one-shot MCP interactionsTests, fixtures, captures, deterministic assertion semantics, and CI reports remain limited or external
Official MCP ConformanceProtocol compliance/interoperabilityRepeatable client/server protocol checksEvolving protocol suite rather than application-domain regression testing
MCP TypeScript SDKSDK and in-memory testingTyped clients and in-process transport testingLanguage/repository coupled; in-memory tests do not cover deployed transport boundaries
MCP Python SDKSDK and in-memory testingPython integration tests and memory helpersLanguage/repository coupled rather than a universal black-box suite
FastMCP testingPython framework testingConvenient deterministic pytest/in-process testingFramework-specific and can bypass subprocess, HTTP, auth, proxy, and deployment behavior
mcptoolsCLI probingShell-friendly list/call/read operationsAssertion model, scenarios, reports, fixtures, and coverage are external
Promptfoo MCP providerBehavioral/agentic evaluationTool selection and response-quality evaluation with reportsLLM/model behavior can be probabilistic; not wire or deterministic application testing
mcp-vibetestAgentic semantic evaluationRealistic task-completion checksProbabilistic and complementary to hard deterministic gates
mcp-scanSecurity scanningTool-poisoning and prompt/config risk detectionSpecialized security gate, not functional regression testing
Cisco MCP ScannerSecurity scanningMCP-focused security analysisSpecialized scanner rather than behavior/protocol test runner
MCPJam InspectorInteractive workbenchFriendly exploratory MCP developmentPrimarily interactive; deterministic CI should not be assumed without specific current support

Project details and release status change quickly. Pin versions and verify current primary documentation before making compatibility claims.

Positioning

Category

Deterministic MCP application testing.

One-line message

MCP Rigor is the open-source test runner for repeatable MCP server behavior across stdio and Streamable HTTP.

Memorable message

Playwright for MCP: author once, run the same scenarios locally and in CI.

Naming rationale

“MCP Rigor” uses established developer-tool terminology, clearly describes the project’s scope, and remains independent from any specific MCP server or vendor brand. Registry and trademark availability should still be verified before publication.

Durable differentiation

  1. Application workflows, not only probes: multi-step scenarios, captures, fixtures, setup/teardown, and side-effect verification.
  2. Real boundaries: subprocess lifecycle, stdio framing, Streamable HTTP, auth, TLS/proxy, reconnect, and cleanup.
  3. Stable CI contract: deterministic exit codes, JSON/JUnit/SARIF, traces, artifacts, filtering, and policy-controlled retries.
  4. MCP-aware assertions: tools/resources/prompts, errors, notifications, progress, pagination, cancellation, capability/version matrices, and subscriptions.
  5. Declarative and programmable: safe YAML/JSON for most cases plus a typed API and explicit plugin model for advanced projects.
  6. Layered oracles: deterministic assertions as hard gates; optional semantic/LLM evaluation as a clearly labeled soft layer.
  7. Record to regression: capture an exploratory interaction, sanitize secrets/dynamic values, and generate a checked-in test.
  8. Evidence, not certification: run or aggregate exact official conformance profiles without making blanket compliance claims.

What not to become

Launch wedge

Target teams that have one or more MCP servers in CI and currently use shell scripts, Inspector screenshots, ad hoc SDK tests, or manual checks. The first compelling demo should:

  1. run the same calculator test against stdio and HTTP targets;
  2. show a clear structural diff for a broken tool response;
  3. capture a value and reuse it in a later step;
  4. verify a protocol error and timeout;
  5. export JUnit in GitHub Actions;
  6. prove the child process is cleaned up after failure.

Suggested public narrative

“The MCP ecosystem already has excellent tools to inspect a server and increasingly strong official protocol conformance checks. MCP Rigor addresses a different question: does your server still perform the domain behaviors your users depend on? It turns those behaviors into deterministic scenarios that run over real transports on every pull request.”