Chrome DevTools MCP: live evidence, explicit scope and verifiable browser work
Build a browser-evidence ledger before automating a full debugging workflow
Track targets, observations, hypotheses and verified outcomes in a read-only prototype, keeping sensitive evidence and automatic actions out of scope.
What you will learn
- Store observations separately from conclusions
- Require evidence for completion
- Add automation only after the read-only model helps
Before you start
- Basic browser console and network concepts
- Understanding of CLI arguments and local client configuration
Plan a scoped browser investigation and distinguish interface defaults, tool success and end-to-end acceptance.
Key takeaways
- Observations and causal hypotheses are different records.
- A passing narrow test should retain its narrow scope.
- Read-only evidence organization can precede automation.
Store observations separately from conclusions
A useful learning extension records the selected page, timestamp, page revision, observation type and result. Keep the hypothesis in a separate field with supporting and conflicting evidence. A console error and a broken screenshot may be related, but their co-occurrence does not establish causation.
Use synthetic records in the first prototype. The ledger is our proposal, not an existing Chrome DevTools MCP feature. Do not populate it with private browsing history or real session cookies just to make a demo look convincing.
Require evidence for completion
A diagnostic finding should name what was observed, the proposed explanation and the acceptance check still needed. A successful click or trace collection can complete one step without resolving the defect. Keep unknowns visible rather than giving every record a success badge.
The fourteen serializer fixture cases provide a model for precise evidence labeling: they prove an isolated function property, not a working daemon or browser. The ledger should preserve that scope instead of turning any test pass into end-to-end confidence.
Add automation only after the read-only model helps
Begin with a table and links to approved captures. A timeline can make the order of navigation, observation and verification clearer; three-dimensional decoration does not improve this relationship. Define retention and redaction before importing real diagnostic material.
A later integration may gather permitted observations, but replaying actions, restarting a daemon and publishing findings should remain separate authorized steps. This proposal opens no browser, submits no forms and creates no external monitoring or scheduled job.
Decision guide
| Criterion | Option A | Option B |
|---|---|---|
| Best when | You need predictable behavior and easy auditing | You need adaptive optimization and have reliable telemetry |
| Main risk | May leave performance on the table | Can become difficult to explain or debug |
Implementation steps
- 1
Define synthetic target and observation fields.
- 2
Separate hypothesis from supporting evidence.
- 3
Record missing acceptance checks explicitly.
- 4
Validate retention and authority before live integration.
Copy-ready example
{
"proposal": true,
"targetPageId": null,
"observation": "synthetic console error",
"hypothesis": null,
"causationVerified": false,
"acceptanceCheck": "not run",
"browserOpened": false,
"monitorScheduled": false
}Frequently asked questions
Does the ledger automatically fix browser defects?
No. It organizes evidence and pending verification.
Will this proposal start monitoring real browsing?
No. It is a synthetic read-only design with no browser or schedule created.
Sources
- Chrome DevTools MCP / README.mdSource checked 2026-09-14
- Chrome DevTools MCP / docs/configuration.mdSource checked 2026-09-14
- Chrome DevTools MCP / src/daemon/utils.tsSource checked 2026-09-14