An independent field guide to agent quirksby Polytomic
A puzzled pixel computer with a question mark on its blue screen

What the Harness?

Same protocol. Different behavior.

Test 001: The double payload

MCP output

Does your harness pass tool output once, or twice?

Why it matters

We don’t want to flood the model with duplicate tokens. MCP tools can return the same data as text and structured output; the harness should pass it to the model once.

This test puts a different random marker in each field and asks the agent what it sees. Reporting both markers reveals duplication. Two extra probes check that text-only and structured-only responses still get through.

What counts as a pass?

Both fields sent

Pass: uses structured output
Warn: falls back to text
Fail: duplicate, missing, or unexpected output

Text only sent

Pass: only text gets through
Fail: missing or unexpected output

Structured only sent

Pass: only structured output gets through
Fail: missing or unexpected output

Our rubric: prefer one structured payload, avoid duplication, preserve single-field results. This preference is not a claim about what the MCP standard mandates.

Observed results

Run history ↗

Latest observed run per harness · Alphabetical · Expand to compare MCP client versions

Scroll the table sideways to compare all three responses →

Test 001 observations. Each result applies to one recorded session. Desired outcomes are defined in the rubric above.
HarnessResponse sent by the probeEvidence
Both fieldscontent + structuredContentText onlycontentStructured onlystructuredContent
Amp CLILatest observed · MCP client version 1Run: Oct 5, 2026MCP client: amp-thread-actorPassUses structured outputPassPassView ↗
Amp via OrbLatest observed · MCP client version 1Run: Nov 10, 2026MCP client: amp-thread-actorPassUses structured outputPassPassView ↗
ChatGPT Desktop (Scheduled Cloud Task)Latest observed · MCP client version 1.0.0Run: Nov 9, 2026MCP client: openai-mcp (Codex)FailDuplicate outputPassPassView ↗
ChatGPT Desktop (Scheduled Local Task)Latest observed · MCP client version 0.167.0-alpha.3.2Run: Nov 9, 2026Same results across 5 observationsFailDuplicate outputPassPassView ↗
MCP client version 0.167.0-alpha.3.2Latest observedMCP client: codex-mcp-clientRun: Nov 9, 2026FailDuplicate outputPassPassView ↗
MCP client version 0.166.0-alpha.5.1MCP client: codex-mcp-clientRun: Nov 2, 2026FailDuplicate outputPassPassView ↗
MCP client version 0.164.0-alpha.11.1MCP client: codex-mcp-clientRun: Oct 26, 2026FailDuplicate outputPassPassView ↗
MCP client version 0.163.0-alpha.10.2MCP client: codex-mcp-clientRun: Oct 19, 2026FailDuplicate outputPassPassView ↗
MCP client version 0.162.0-alpha.17.2MCP client: codex-mcp-clientRun: Oct 13, 2026FailDuplicate outputPassPassView ↗
ChatGPT Desktop CodexLatest observed · MCP client version 0.160.0Run: Oct 5, 2026MCP client: codex-mcp-clientFailDuplicate outputPassPassView ↗
Claude Code (Claude desktop app, Code tab)Latest observed · MCP client version 2.1.286Run: Oct 5, 2026Same results across 2 observationsPassUses structured outputPassPassView ↗
MCP client version 2.1.286Latest observedMCP client: claude-codeRun: Oct 5, 2026PassUses structured outputPassPassView ↗
MCP client version 1.0.0MCP client: Anthropic/ClaudeAIRun: Oct 5, 2026PassUses structured outputPassPassView ↗
Claude Code CLILatest observed · MCP client version 2.1.314Run: Nov 2, 2026Same results across 4 observationsPassUses structured outputPassPassView ↗
MCP client version 2.1.314Latest observedMCP client: claude-codeRun: Nov 2, 2026PassUses structured outputPassPassView ↗
MCP client version 2.1.309MCP client: claude-codeRun: Oct 26, 2026PassUses structured outputPassPassView ↗
MCP client version 2.1.296MCP client: claude-codeRun: Oct 13, 2026PassUses structured outputPassPassView ↗
MCP client version 2.1.289MCP client: claude-codeRun: Oct 5, 2026PassUses structured outputPassPassView ↗
Claude DesktopLatest observed · MCP client version 1.0.0Run: Oct 5, 2026MCP client: Anthropic/ClaudeAIWarnFalls back to textPassPassView ↗
Claude Desktop (Scheduled Cloud Task)Latest observed · MCP client version 2.1.310Run: Oct 26, 2026Same results across 2 observationsPassUses structured outputPassPassView ↗
MCP client version 2.1.310Latest observedMCP client: claude-codeRun: Oct 26, 2026PassUses structured outputPassPassView ↗
MCP client version 2.1.296MCP client: claude-codeRun: Oct 13, 2026PassUses structured outputPassPassView ↗
Claude Desktop (Scheduled Local Task)Latest observed · MCP client version 2.1.319Run: Nov 9, 2026Same results across 5 observationsPassUses structured outputPassPassView ↗
MCP client version 2.1.319Latest observedMCP client: claude-codeRun: Nov 9, 2026PassUses structured outputPassPassView ↗
MCP client version 2.1.316MCP client: claude-codeRun: Nov 2, 2026PassUses structured outputPassPassView ↗
MCP client version 2.1.310MCP client: claude-codeRun: Oct 26, 2026PassUses structured outputPassPassView ↗
MCP client version 2.1.303MCP client: claude-codeRun: Oct 19, 2026PassUses structured outputPassPassView ↗
MCP client version 2.1.296MCP client: claude-codeRun: Oct 13, 2026PassUses structured outputPassPassView ↗
Codex CLILatest observed · MCP client version 0.164.0Run: Nov 9, 2026Same results across 3 observationsFailDuplicate outputPassPassView ↗
MCP client version 0.164.0Latest observedMCP client: codex-mcp-clientRun: Nov 9, 2026FailDuplicate outputPassPassView ↗
MCP client version 0.162.1MCP client: codex-mcp-clientRun: Oct 19, 2026FailDuplicate outputPassPassView ↗
MCP client version 0.160.1MCP client: codex-mcp-clientRun: Oct 5, 2026FailDuplicate outputPassPassView ↗
Cursor Agent CLILatest observed · MCP client version 1.0.0Run: Nov 9, 2026MCP client: CursorWarnFalls back to textPassFailNo outputView ↗
Cursor DesktopLatest observed · MCP client version 1.0.0Run: Oct 5, 2026MCP client: cursor-vscodeWarnFalls back to textPassFailNo outputView ↗
Grok BotLatest observed · MCP client version 0.1.0Run: Nov 9, 2026MCP client: connectors-managerFailDuplicate outputPassPassView ↗
Grok CLILatest observed · MCP client version 1.0.70Run: Nov 9, 2026Same results across 5 observationsFailDuplicate outputPassPassView ↗
MCP client version 1.0.70Latest observedMCP client: grok-shell-what-the-harnessRun: Nov 9, 2026FailDuplicate outputPassPassView ↗
MCP client version 1.0.62MCP client: grok-shell-what-the-harnessRun: Oct 26, 2026FailDuplicate outputPassPassView ↗
MCP client version 1.0.60MCP client: grok-shell-what-the-harnessRun: Oct 19, 2026FailDuplicate outputPassPassView ↗
MCP client version 1.0.52MCP client: grok-shell-what-the-harnessRun: Oct 13, 2026FailDuplicate outputPassPassView ↗
MCP client version 1.0.46MCP client: grok-shell-what-the-harnessRun: Oct 5, 2026FailDuplicate outputPassPassView ↗
OpenAI Agents APILatest observed · MCP client version 1.0.0Run: Oct 5, 2026MCP client: openai-mcp (Codex)FailDuplicate outputPassPassView ↗
piLatest observed · MCP client version 1.0.3Run: Oct 5, 2026MCP client: piFailDuplicate outputPassPassView ↗
Pass Desired outputWarn Text fallbackFail Duplicate, missing,
or unexpected output

Verdicts use reported markers. Observations describe individual sessions, not inferred model inputs.

Test 002: Protocol version

Connection metadata

Which MCP version did this connection use?

The MCP version used in this run. Green means the latest version tested here (2026-07-28); yellow means an older version. Full connection details are in the run evidence.

HarnessProtocol versionEvidence
Amp CLILatest observed · MCP client version 1Run: Oct 5, 2026MCP client: amp-thread-actorOlder protocol2025-06-18View ↗
Amp via OrbLatest observed · MCP client version 1Run: Nov 10, 2026MCP client: amp-thread-actorLatest2026-07-28View ↗
ChatGPT Desktop (Scheduled Cloud Task)Latest observed · MCP client version 1.0.0Run: Nov 9, 2026MCP client: openai-mcp (Codex)Latest2026-07-28View ↗
ChatGPT Desktop (Scheduled Local Task)Latest observed · MCP client version 0.167.0-alpha.3.2Run: Nov 9, 2026Same results across 5 observationsOlder protocol2025-06-18View ↗
MCP client version 0.167.0-alpha.3.2Latest observedMCP client: codex-mcp-clientRun: Nov 9, 2026Older protocol2025-06-18View ↗
MCP client version 0.166.0-alpha.5.1MCP client: codex-mcp-clientRun: Nov 2, 2026Older protocol2025-06-18View ↗
MCP client version 0.164.0-alpha.11.1MCP client: codex-mcp-clientRun: Oct 26, 2026Older protocol2025-06-18View ↗
MCP client version 0.163.0-alpha.10.2MCP client: codex-mcp-clientRun: Oct 19, 2026Older protocol2025-06-18View ↗
MCP client version 0.162.0-alpha.17.2MCP client: codex-mcp-clientRun: Oct 13, 2026Older protocol2025-06-18View ↗
ChatGPT Desktop CodexLatest observed · MCP client version 0.160.0Run: Oct 5, 2026MCP client: codex-mcp-clientOlder protocol2025-06-18View ↗
Claude Code (Claude desktop app, Code tab)Latest observed · MCP client version 2.1.286Run: Oct 5, 2026Same results across 2 observationsLatest2026-07-28View ↗
MCP client version 2.1.286Latest observedMCP client: claude-codeRun: Oct 5, 2026Latest2026-07-28View ↗
MCP client version 1.0.0MCP client: Anthropic/ClaudeAIRun: Oct 5, 2026Latest2026-07-28View ↗
Claude Code CLILatest observed · MCP client version 2.1.314Run: Nov 2, 2026Same results across 4 observationsLatest2026-07-28View ↗
MCP client version 2.1.314Latest observedMCP client: claude-codeRun: Nov 2, 2026Latest2026-07-28View ↗
MCP client version 2.1.309MCP client: claude-codeRun: Oct 26, 2026Latest2026-07-28View ↗
MCP client version 2.1.296MCP client: claude-codeRun: Oct 13, 2026Latest2026-07-28View ↗
MCP client version 2.1.289MCP client: claude-codeRun: Oct 5, 2026Latest2026-07-28View ↗
Claude DesktopLatest observed · MCP client version 1.0.0Run: Oct 5, 2026MCP client: Anthropic/ClaudeAILatest2026-07-28View ↗
Claude Desktop (Scheduled Cloud Task)Latest observed · MCP client version 2.1.310Run: Oct 26, 2026Same results across 2 observationsLatest2026-07-28View ↗
MCP client version 2.1.310Latest observedMCP client: claude-codeRun: Oct 26, 2026Latest2026-07-28View ↗
MCP client version 2.1.296MCP client: claude-codeRun: Oct 13, 2026Latest2026-07-28View ↗
Claude Desktop (Scheduled Local Task)Latest observed · MCP client version 2.1.319Run: Nov 9, 2026Same results across 5 observationsLatest2026-07-28View ↗
MCP client version 2.1.319Latest observedMCP client: claude-codeRun: Nov 9, 2026Latest2026-07-28View ↗
MCP client version 2.1.316MCP client: claude-codeRun: Nov 2, 2026Latest2026-07-28View ↗
MCP client version 2.1.310MCP client: claude-codeRun: Oct 26, 2026Latest2026-07-28View ↗
MCP client version 2.1.303MCP client: claude-codeRun: Oct 19, 2026Latest2026-07-28View ↗
MCP client version 2.1.296MCP client: claude-codeRun: Oct 13, 2026Latest2026-07-28View ↗
Codex CLILatest observed · MCP client version 0.164.0Run: Nov 9, 2026Same results across 3 observationsOlder protocol2025-06-18View ↗
MCP client version 0.164.0Latest observedMCP client: codex-mcp-clientRun: Nov 9, 2026Older protocol2025-06-18View ↗
MCP client version 0.162.1MCP client: codex-mcp-clientRun: Oct 19, 2026Older protocol2025-06-18View ↗
MCP client version 0.160.1MCP client: codex-mcp-clientRun: Oct 5, 2026Older protocol2025-06-18View ↗
Cursor Agent CLILatest observed · MCP client version 1.0.0Run: Nov 9, 2026MCP client: CursorOlder protocol2025-11-25View ↗
Cursor DesktopLatest observed · MCP client version 1.0.0Run: Oct 5, 2026MCP client: cursor-vscodeOlder protocol2025-11-25View ↗
Grok BotLatest observed · MCP client version 0.1.0Run: Nov 9, 2026MCP client: connectors-managerOlder protocol2025-11-25View ↗
Grok CLILatest observed · MCP client version 1.0.70Run: Nov 9, 2026Same results across 5 observationsLatest2026-07-28View ↗
MCP client version 1.0.70Latest observedMCP client: grok-shell-what-the-harnessRun: Nov 9, 2026Latest2026-07-28View ↗
MCP client version 1.0.62MCP client: grok-shell-what-the-harnessRun: Oct 26, 2026Latest2026-07-28View ↗
MCP client version 1.0.60MCP client: grok-shell-what-the-harnessRun: Oct 19, 2026Latest2026-07-28View ↗
MCP client version 1.0.52MCP client: grok-shell-what-the-harnessRun: Oct 13, 2026Latest2026-07-28View ↗
MCP client version 1.0.46MCP client: grok-shell-what-the-harnessRun: Oct 5, 2026Latest2026-07-28View ↗
OpenAI Agents APILatest observed · MCP client version 1.0.0Run: Oct 5, 2026MCP client: openai-mcp (Codex)Older protocol2025-11-25View ↗
piLatest observed · MCP client version 1.0.3Run: Oct 5, 2026MCP client: piOlder protocol2025-11-25View ↗

Test 003: URL elicitation

Optional interaction

Does the client advertise URL elicitation, and can it complete the flow?

Advertised means the client says it supports URL elicitation. Verified means the browser interaction completed and the tool continued successfully. A client can advertise support while using an older protocol that prevents verification.

HarnessURL support advertisedInteractive verificationEvidence
Amp CLILatest observed · MCP client version 1Run: Oct 5, 2026MCP client: amp-thread-actorNoSkippedView ↗
Amp via OrbLatest observed · MCP client version 1Run: Nov 10, 2026MCP client: amp-thread-actorYesVerifiedView ↗
ChatGPT Desktop (Scheduled Cloud Task)Latest observed · MCP client version 1.0.0Run: Nov 9, 2026MCP client: openai-mcp (Codex)YesNot requestedView ↗
ChatGPT Desktop (Scheduled Local Task)Latest observed · MCP client version 0.167.0-alpha.3.2Run: Nov 9, 2026Same results across 5 observationsYesSkipped: requires newer protocolView ↗
MCP client version 0.167.0-alpha.3.2Latest observedMCP client: codex-mcp-clientRun: Nov 9, 2026YesSkipped: requires newer protocolView ↗
MCP client version 0.166.0-alpha.5.1MCP client: codex-mcp-clientRun: Nov 2, 2026YesSkipped: requires newer protocolView ↗
MCP client version 0.164.0-alpha.11.1MCP client: codex-mcp-clientRun: Oct 26, 2026YesSkipped: requires newer protocolView ↗
MCP client version 0.163.0-alpha.10.2MCP client: codex-mcp-clientRun: Oct 19, 2026YesSkipped: requires newer protocolView ↗
MCP client version 0.162.0-alpha.17.2MCP client: codex-mcp-clientRun: Oct 13, 2026YesSkipped: requires newer protocolView ↗
ChatGPT Desktop CodexLatest observed · MCP client version 0.160.0Run: Oct 5, 2026MCP client: codex-mcp-clientYesSkipped: requires newer protocolView ↗
Claude Code (Claude desktop app, Code tab)Latest observed · MCP client version 2.1.286Run: Oct 5, 2026Results differ across observationsYesDeclinedView ↗
MCP client version 2.1.286Latest observedMCP client: claude-codeRun: Oct 5, 2026YesDeclinedView ↗
MCP client version 1.0.0MCP client: Anthropic/ClaudeAIRun: Oct 5, 2026NoSkippedView ↗
Claude Code CLILatest observed · MCP client version 2.1.314Run: Nov 2, 2026Results differ across observationsYesExpiredView ↗
MCP client version 2.1.314Latest observedMCP client: claude-codeRun: Nov 2, 2026YesExpiredView ↗
MCP client version 2.1.309MCP client: claude-codeRun: Oct 26, 2026YesVerifiedView ↗
MCP client version 2.1.296MCP client: claude-codeRun: Oct 13, 2026YesVerifiedView ↗
MCP client version 2.1.289MCP client: claude-codeRun: Oct 5, 2026YesVerifiedView ↗
Claude DesktopLatest observed · MCP client version 1.0.0Run: Oct 5, 2026MCP client: Anthropic/ClaudeAINoSkippedView ↗
Claude Desktop (Scheduled Cloud Task)Latest observed · MCP client version 2.1.310Run: Oct 26, 2026Same results across 2 observationsYesCancelledView ↗
MCP client version 2.1.310Latest observedMCP client: claude-codeRun: Oct 26, 2026YesCancelledView ↗
MCP client version 2.1.296MCP client: claude-codeRun: Oct 13, 2026YesCancelledView ↗
Claude Desktop (Scheduled Local Task)Latest observed · MCP client version 2.1.319Run: Nov 9, 2026Same results across 5 observationsYesCancelledView ↗
MCP client version 2.1.319Latest observedMCP client: claude-codeRun: Nov 9, 2026YesCancelledView ↗
MCP client version 2.1.316MCP client: claude-codeRun: Nov 2, 2026YesCancelledView ↗
MCP client version 2.1.310MCP client: claude-codeRun: Oct 26, 2026YesCancelledView ↗
MCP client version 2.1.303MCP client: claude-codeRun: Oct 19, 2026YesCancelledView ↗
MCP client version 2.1.296MCP client: claude-codeRun: Oct 13, 2026YesCancelledView ↗
Codex CLILatest observed · MCP client version 0.164.0Run: Nov 9, 2026Same results across 3 observationsYesSkipped: requires newer protocolView ↗
MCP client version 0.164.0Latest observedMCP client: codex-mcp-clientRun: Nov 9, 2026YesSkipped: requires newer protocolView ↗
MCP client version 0.162.1MCP client: codex-mcp-clientRun: Oct 19, 2026YesSkipped: requires newer protocolView ↗
MCP client version 0.160.1MCP client: codex-mcp-clientRun: Oct 5, 2026YesSkipped: requires newer protocolView ↗
Cursor Agent CLILatest observed · MCP client version 1.0.0Run: Nov 9, 2026MCP client: CursorNoSkippedView ↗
Cursor DesktopLatest observed · MCP client version 1.0.0Run: Oct 5, 2026MCP client: cursor-vscodeNoSkippedView ↗
Grok BotLatest observed · MCP client version 0.1.0Run: Nov 9, 2026MCP client: connectors-managerNoSkippedView ↗
Grok CLILatest observed · MCP client version 1.0.70Run: Nov 9, 2026Results differ across observationsYesIncompleteView ↗
MCP client version 1.0.70Latest observedMCP client: grok-shell-what-the-harnessRun: Nov 9, 2026YesIncompleteView ↗
MCP client version 1.0.62MCP client: grok-shell-what-the-harnessRun: Oct 26, 2026YesIncompleteView ↗
MCP client version 1.0.60MCP client: grok-shell-what-the-harnessRun: Oct 19, 2026YesExpiredView ↗
MCP client version 1.0.52MCP client: grok-shell-what-the-harnessRun: Oct 13, 2026YesIncompleteView ↗
MCP client version 1.0.46MCP client: grok-shell-what-the-harnessRun: Oct 5, 2026YesIncompleteView ↗
OpenAI Agents APILatest observed · MCP client version 1.0.0Run: Oct 5, 2026MCP client: openai-mcp (Codex)NoSkippedView ↗
piLatest observed · MCP client version 1.0.3Run: Oct 5, 2026MCP client: piNoSkippedView ↗

Each harness shows its latest observed run. Missing tests are labelled explicitly. Expand versions or open the run history for earlier evidence.

Test 004: Resource reading

MCP resources

Can the agent read a resource, including one linked by a tool?

Two resources contain independent random markers. Green means the server recorded a read and the agent reported the exact marker. Yellow means it was not read or its marker was not reported. Red means an incorrect marker was reported.

HarnessDirect resourceTool resource linkEvidence
Amp CLILatest observed · MCP client version 1Run: Oct 5, 2026MCP client: amp-thread-actorNot readNot readView ↗
Amp via OrbLatest observed · MCP client version 1Run: Nov 10, 2026MCP client: amp-thread-actorNot readNot readView ↗
ChatGPT Desktop (Scheduled Cloud Task)Latest observed · MCP client version 1.0.0Run: Nov 9, 2026MCP client: openai-mcp (Codex)Not tested in this runNot tested in this runView ↗
ChatGPT Desktop (Scheduled Local Task)Latest observed · MCP client version 0.167.0-alpha.3.2Run: Nov 9, 2026Same results across 5 observationsRead successfullyRead successfullyView ↗
MCP client version 0.167.0-alpha.3.2Latest observedMCP client: codex-mcp-clientRun: Nov 9, 2026Read successfullyRead successfullyView ↗
MCP client version 0.166.0-alpha.5.1MCP client: codex-mcp-clientRun: Nov 2, 2026Read successfullyRead successfullyView ↗
MCP client version 0.164.0-alpha.11.1MCP client: codex-mcp-clientRun: Oct 26, 2026Read successfullyRead successfullyView ↗
MCP client version 0.163.0-alpha.10.2MCP client: codex-mcp-clientRun: Oct 19, 2026Read successfullyRead successfullyView ↗
MCP client version 0.162.0-alpha.17.2MCP client: codex-mcp-clientRun: Oct 13, 2026Read successfullyRead successfullyView ↗
ChatGPT Desktop CodexLatest observed · MCP client version 0.160.0Run: Oct 5, 2026MCP client: codex-mcp-clientRead successfullyRead successfullyView ↗
Claude Code (Claude desktop app, Code tab)Latest observed · MCP client version 2.1.286Run: Oct 5, 2026Results differ across observationsRead successfullyRead successfullyView ↗
MCP client version 2.1.286Latest observedMCP client: claude-codeRun: Oct 5, 2026Read successfullyRead successfullyView ↗
MCP client version 1.0.0MCP client: Anthropic/ClaudeAIRun: Oct 5, 2026Not readNot readView ↗
Claude Code CLILatest observed · MCP client version 2.1.314Run: Nov 2, 2026Same results across 4 observationsRead successfullyRead successfullyView ↗
MCP client version 2.1.314Latest observedMCP client: claude-codeRun: Nov 2, 2026Read successfullyRead successfullyView ↗
MCP client version 2.1.309MCP client: claude-codeRun: Oct 26, 2026Read successfullyRead successfullyView ↗
MCP client version 2.1.296MCP client: claude-codeRun: Oct 13, 2026Read successfullyRead successfullyView ↗
MCP client version 2.1.289MCP client: claude-codeRun: Oct 5, 2026Read successfullyRead successfullyView ↗
Claude DesktopLatest observed · MCP client version 1.0.0Run: Oct 5, 2026MCP client: Anthropic/ClaudeAIRead successfullyRead successfullyView ↗
Claude Desktop (Scheduled Cloud Task)Latest observed · MCP client version 2.1.310Run: Oct 26, 2026Same results across 2 observationsRead successfullyRead successfullyView ↗
MCP client version 2.1.310Latest observedMCP client: claude-codeRun: Oct 26, 2026Read successfullyRead successfullyView ↗
MCP client version 2.1.296MCP client: claude-codeRun: Oct 13, 2026Read successfullyRead successfullyView ↗
Claude Desktop (Scheduled Local Task)Latest observed · MCP client version 2.1.319Run: Nov 9, 2026Same results across 5 observationsRead successfullyRead successfullyView ↗
MCP client version 2.1.319Latest observedMCP client: claude-codeRun: Nov 9, 2026Read successfullyRead successfullyView ↗
MCP client version 2.1.316MCP client: claude-codeRun: Nov 2, 2026Read successfullyRead successfullyView ↗
MCP client version 2.1.310MCP client: claude-codeRun: Oct 26, 2026Read successfullyRead successfullyView ↗
MCP client version 2.1.303MCP client: claude-codeRun: Oct 19, 2026Read successfullyRead successfullyView ↗
MCP client version 2.1.296MCP client: claude-codeRun: Oct 13, 2026Read successfullyRead successfullyView ↗
Codex CLILatest observed · MCP client version 0.164.0Run: Nov 9, 2026Same results across 3 observationsRead successfullyRead successfullyView ↗
MCP client version 0.164.0Latest observedMCP client: codex-mcp-clientRun: Nov 9, 2026Read successfullyRead successfullyView ↗
MCP client version 0.162.1MCP client: codex-mcp-clientRun: Oct 19, 2026Read successfullyRead successfullyView ↗
MCP client version 0.160.1MCP client: codex-mcp-clientRun: Oct 5, 2026Read successfullyRead successfullyView ↗
Cursor Agent CLILatest observed · MCP client version 1.0.0Run: Nov 9, 2026MCP client: CursorRead successfullyRead successfullyView ↗
Cursor DesktopLatest observed · MCP client version 1.0.0Run: Oct 5, 2026MCP client: cursor-vscodeRead successfullyRead successfullyView ↗
Grok BotLatest observed · MCP client version 0.1.0Run: Nov 9, 2026MCP client: connectors-managerNot readNot readView ↗
Grok CLILatest observed · MCP client version 1.0.70Run: Nov 9, 2026Same results across 5 observationsNot readNot readView ↗
MCP client version 1.0.70Latest observedMCP client: grok-shell-what-the-harnessRun: Nov 9, 2026Not readNot readView ↗
MCP client version 1.0.62MCP client: grok-shell-what-the-harnessRun: Oct 26, 2026Not readNot readView ↗
MCP client version 1.0.60MCP client: grok-shell-what-the-harnessRun: Oct 19, 2026Not readNot readView ↗
MCP client version 1.0.52MCP client: grok-shell-what-the-harnessRun: Oct 13, 2026Not readNot readView ↗
MCP client version 1.0.46MCP client: grok-shell-what-the-harnessRun: Oct 5, 2026Not readNot readView ↗
OpenAI Agents APILatest observed · MCP client version 1.0.0Run: Oct 5, 2026MCP client: openai-mcp (Codex)Not readNot readView ↗
piLatest observed · MCP client version 1.0.3Run: Oct 5, 2026MCP client: piRead successfullyRead successfullyView ↗

Run it yourself

  1. Add the public MCP connection.
  2. Begin a run to record protocol and capabilities, then call the output probes and read both test resources.
  3. Optionally verify URL elicitation, then submit the exact markers you see for review.
https://what-the-harness.polytomic.com/mcp

No client authorization is required. Connection guide. The prompt runs all four tests with optional human verification. To skip the interaction, set verify_url_elicitation to false. The same endpoint supports standalone manual probes.

Read the test prompt

Use the What the Harness MCP server. Call begin_run with verify_url_elicitation: true and test_resources: true; if its schema requires begin_key, generate a fresh UUID for this run, keep it private, and reuse it for start retries. Supply the harness and model only if explicitly known, otherwise leave them null. Save the returned run_id, run_token, and client-declared identity. Keep resource URIs and run_token private; omit it from the final report and evidence. Call probe_both, probe_text_only, and probe_structured_only once each with that run_id and run_token. For each initial normal tool call, record the exact probe_id, text_marker, and structured_marker visible in its response. Use null for every value not visible; do not infer or invent markers. Do not use raw-response APIs (such as .raw()), repeat successful probes, or perform additional inspection to recover missing output markers. Report only what the initial normal calls exposed. Discover MCP resources and resource templates if the client makes them available. Read the private resources.direct URI from begin_run through MCP resources/read and record its exact resource_marker. Call probe_resource_link and read the URI in its resource_link through MCP; record that resource_marker separately. If a resource cannot be read or its marker is not visible, report null. Never fetch these URIs over HTTP, inspect server files, or infer markers. Call probe_url_elicitation with the same run_id and run_token. If it presents a URL interaction, pause for the human to complete it through the client UI; do not fetch or complete the page yourself. After the human completes it, retry probe_url_elicitation with the same credentials until it returns the verification result. If support is not advertised or the protocol is unsupported, record the skip. If the human declines or cancels, record that outcome. Call submit_result with the run_id, run_token, and all three output reports, plus resource_reports containing direct and linked markers (or null), using those exact field names. If submission times out, retry with the same run_id, run_token, and unchanged reports. If the MCP connection reconnects, keep using the original run_id and run_token. Call result_status with the same credentials. Report the receipt, publication state, protocol version, declared URL support, verification status, and the output and resource markers you saw; say “not visible” for null values. Do not fetch the endpoint or website separately, and do not start another run to recover a missing marker.

How to read these results
  • Reporting an exact random marker demonstrates visibility.
  • A missing marker alone does not establish the cause.
  • Results apply to the recorded client version and session.

Test method and controls

Your agent here.

Run the probe and save the exact report, client/version, model, date, and evidence. Results are reviewed before publication.

Your agent submits through the public MCP connection. Connection guide →

Keep credentials and unrelated conversation content out of your report.