Let the agent verify the app it just changed.

AI coding agents can write a checkout flow quickly. The missing step is using the application and checking the outcome. Molar gives agents a CLI and MCP surface for authorized browser tasks, observations, traces, and integration fixtures.

Get started Read the docs

Code → browser → evidenceExample workflow

Coding agent

The login form is implemented.
Check that the dashboard loads after sign-in.

molar verify "login works"
--url http://localhost:3000 --json

Molar

Open the browser. Execute the flow.
Return the observed outcome and trace.

InspectFixVerify again

One command after code

Run molar verify "login works" --url http://localhost:3000 --json and return a structured pass or failure for the next agent turn.

MCP native

Connect an agent to https://api.molar.it/mcp or run the local stdio server through molar mcp serve, subject to the deployed client surface.

Controlled autonomy

External targets require authorization, and financial or irreversible browser actions pause for confirmation.

Useful output for the next turn

Task IDs, status, rationale, trace references, live view fields when available, and artifact manifests give an agent enough context to diagnose and rerun.

JSON output
Task polling
Scoped browser and identity tools

Go to documentation

Browser taskIllustrative workflow

billing.acme.test

Browser taskFind August invoice Search invoices

Aug 2026 Paid

Documentinvoice_august_2026.pdf

Downloaded

A verification loop with control

Local browser verification is the shortest path. Hosted runs use scoped credentials and target authorization, with beta availability varying by workflow.

Talk through your use case

Live evidenceIllustrative workflow

checkout.acme.test / confirmation

Order confirmedReceipt sent to test inboxpayment_intent.succeeded

Run trace 00:08—00:10

Open checkoutScreenshot captured200

Confirm paymentPOST /webhooks/payment200

Receive receiptTest inbox matchedready

Network and console evidence attached

A few things
you might be wondering.

Have a specific workflow in mind?
Let’s talk it through

Can an agent click a payment button automatically?

It can act on an authorized target, but financial or irreversible steps require the configured confirmation flow.

Can an agent use a real user's login?

Login uses an exact origin saved identity or a referenced credential, with connector controlled OTP support; inline secrets and personal inbox access are not the workflow.

What happens when verification fails?

The CLI can emit a structured failure with a hint and evidence references so the coding agent can inspect the cause and rerun.

Give your agent a verifier.

Choose a concrete workflow. We’ll help you get the right setup for your application.

Get started Talk to the team