Gemini CLI MCP for Bug Reports

Gemini CLI is a terminal AI agent with file and shell tooling. That makes it useful for moving between a bug report, repository files, tests, and command output without switching to a separate report viewer. Gemini CLI documentation confirms the terminal-agent model and its file and shell capabilities.

The useful division is simple:

• clip.qa records screen behavior and can provide browser console, network, screenshot, and video evidence for supported captures.
• Gemini CLI interprets that evidence and inspects the local repository.
• The MCP connection lets Gemini retrieve report data during the task.
• You decide what the agent may change, which tools it may use, and whether a proposed fix has actually passed validation.

For a broader LLM bug-reporting approach, see LLM-ready bug reports. To improve the evidence request, use how to describe a bug to AI.

Gemini CLI supports local stdio and remote HTTP or SSE MCP servers. For clip.qa, use the documented remote HTTP endpoint and complete its OAuth flow. MCP tools are available across clip.qa tiers, including Free. Gemini CLI MCP server documentation confirms supported transports, OAuth, and the commands below.

Configure the server with this exact command:

gemini mcp add --transport http clip-qa https://mcp.clip.qa/mcp

Then follow these steps:

1. Run gemini mcp list and confirm the clip-qa server appears with the expected HTTP transport.
2. Start Gemini CLI and run /mcp auth clip-qa interactively.
3. Complete the browser-based OAuth 2.1 PKCE authorization flow as the intended clip.qa user.
4. Ask Gemini to list discovered tools without making a report request yet. Confirm that listReports and getDetails are exposed with their exact remote names.
5. Test listReports, select one known report, and retrieve it with getDetails.
6. Keep write-capable tools separate until you need createComment, linkFix, or linkFixAttempt.
7. Start a debugging task with one report identifier and explicit acceptance criteria.

Do not approve every command or tool by reflex. Read-only permissions should come first, and write tools should be configured separately according to the task’s needs. The documented setup above is based on Gemini CLI and clip.qa documentation; it is not an end-to-end integration test.

Authentication can succeed while the agent still lacks a useful retrieval request. Verify each layer separately. First, gemini mcp list confirms configuration. Second, /mcp auth clip-qa confirms user authorization. Third, tool discovery shows which remote functions Gemini can call.

Use a small verification prompt:

Use the clip-qa MCP server only for this check.

1. List the available clip.qa tools without editing files or running shell commands.
2. Call listReports.
3. Do not open a report yet.
4. Show the server name, transport, discovered tool names, and the number of returned reports.

After that passes, test one known report:

Call listReports and return only the report IDs that are visible to the
authenticated account. Do not expose unrelated report content. I will choose
one ID for a separate getDetails request.

Then retrieve the selected report. This confirms retrieval without turning the first repository prompt into a broad data dump. Reports stay private until explicitly shared.

Use a concrete example that separates an accepted server request from an incorrect interface update. Assume a product page saves successfully, but the displayed quantity remains stale after returning to the cart:

Investigate clip.qa report <REPORT_ID> in the current repository.

Observed: The save request succeeds, the confirmation toast appears, but the cart
still displays the previous quantity.
Expected: The visible quantity matches the quantity confirmed by the server.
Unchanged: Saved payload shape, payment flow, and desktop layout.

First inspect report details, screenshots, video, console output, and network
requests. State whether the evidence shows a failed request or a successful request
followed by stale UI state. Locate the state-update path and propose the smallest
fix and a regression test. Do not edit files until you state your hypothesis and
planned file changes.

This structure gives the agent a falsifiable hypothesis. It also reduces the chance that Gemini will “fix” a toast while missing the state synchronization problem. For more prompting guidance, see screen recording to bug report.

Ask Gemini to build an evidence chain before changing code. For the stale-quantity example, the chain should contain these checks:

1. Identify the user action that triggers the request.
2. Inspect the response status and returned quantity.
3. Compare the response quantity with the quantity shown in the interface.
4. Find where successful responses update client state.
5. Check whether the component derives quantity from stale props, an outdated closure, or a cache that was never refreshed.
6. Write a test that fails before the change and passes after it.
7. Retest the same flow in the environment where the report was captured.

A network request can succeed while the UI remains incorrect. That observation narrows the investigation but does not identify the exact code defect by itself. The agent should cite the relevant console message, response, screenshot, or recorded transition before claiming a root cause.

clip.qa’s iOS and Android apps and Chrome extension are in beta; request access before planning a capture. Video analysis runs in the cloud and produces a draft for a person to review. Console and network capture applies to the browser extension, not arbitrary native mobile apps. This setup follows the published documentation; it has not been tested end to end here. See the clip.qa MCP page for product details.

The same evidence limits apply to platform claims. Browser console and network data can support a browser investigation. They do not expose arbitrary native mobile-app internals, and a terminal-based cloud or local agent cannot prove behavior on a physical phone unless it runs in that original environment.

Use the smallest tool set that answers the current question. The clip.qa remote tool set includes listReports, getDetails, getNetworkRequests, getConsoleLogs, getScreenshots, getVideoTranscript, getReportDiff, createComment, linkFix, and linkFixAttempt.

A read-only investigation can usually begin with:

• listReports to locate the correct record.
• getDetails to read its summary and metadata.
• getScreenshots to inspect visible states.
• getVideoTranscript to locate the relevant action sequence.
• getNetworkRequests and getConsoleLogs to check browser-side behavior.
• getReportDiff to compare two reports.

Reserve createComment for a durable review note, linkFix for associating a code change, and linkFixAttempt for an explicit validation attempt. A linked attempt should move through pending, succeeded, or failed based on evidence. Do not mark it succeeded because Gemini reports that tests passed in a different environment.

Treat Gemini’s output as a proposal until validation confirms it. Check that the diff addresses the stale quantity rather than adding an unrelated refresh call. Verify that the test asserts the user-visible quantity after a successful response. Confirm that request payloads, payment behavior, and desktop presentation remain unchanged.

Then reproduce the original flow in the matching browser environment. If the report concerns a native app, obtain a native test in the relevant build and device state. Terminal tests can validate logic, but they do not replace that retest.

clip.qa generates a draft report from cloud video analysis and has a person review it. It does not guarantee diagnostic accuracy or apply an autonomous fix. Gemini CLI’s role is to use approved evidence and tools; your review determines whether the reasoning and change are sound.

Sign up free to capture evidence for an AI-assisted debugging workflow.

• Gemini CLI documentation — Confirms Gemini CLI as a terminal AI agent with file and shell tooling. Checked 2026-11-05.
• Gemini CLI MCP server documentation — Confirms local stdio and remote HTTP/SSE support, OAuth, mcpServers settings, gemini mcp add --transport http, gemini mcp list, and /mcp auth <server-name>. Checked 2026-11-05.
• clip.qa MCP — Product reference for the remote MCP setup and report workflow. Checked 2026-11-05.