# Browserbase docs: AI Agent Readiness Score 92.5% (A-)

**92.5% · 74/80 · AI Agent Readiness Score · 24/30 reading points · 50/50 agent surface points**

Browserbase ships the complete agent-facing documentation stack. Its main miss is two live pages giving different free-plan limits.

- Tested: 2026-08-19
- Verified: 2026-08-19
- Published: 2026-08-19
- Battery: v1
- Scoring: reading 30 pts · surface 50 pts
- Docs: https://docs.browserbase.com

Three AI models, GPT 5.6 Sol, Claude Opus 5, and DeepSeek v4 Flash, each read Browserbase’s public documentation independently and attempted five first-hour developer jobs: start a browser session, find the exact limits, recover from a 429, authenticate end to end, and use the Node.js SDK.

Browserbase did not expose a product-webhook interface during preflight, so the fixed authentication substitute replaced the webhook task. No accounts, API calls, or code execution were used. Every verdict came from the public pages.

## Agent surface checks · 50/50

| Check | Verdict | Points |
| --- | --- | --- |
| llms.txt | PASS | 10 |
| llms-full.txt | PASS | 10 |
| Markdown mirror | PASS | 10 |
| MCP server | PASS | 10 |
| Docs AI | PASS | 10 |

## The Reading Test · 24/30

| Task | GPT 5.6 Sol | Opus 5 | DeepSeek v4F | Consensus |
| --- | --- | --- | --- | --- |
| Start a browser session | PASS | PASS | PASS | PASS |
| Find the exact limits | PARTIAL | PARTIAL | PARTIAL | PARTIAL |
| Recover from a 429 | PARTIAL | PARTIAL | PASS | PARTIAL |
| Authenticate end to end | PASS | PASS | PASS | PASS |
| Use the Node.js SDK | PASS | PARTIAL | PASS | PASS |

Docs platform: Mintlify (unscored) · verified 2026-08-19

## What the docs get right

- **The complete agent-facing surface.** Browserbase publishes llms.txt, llms-full.txt, page-level markdown, docs AI, and a live documentation MCP server.
- **A quickstart with no missing step.** The Playwright path runs from the dashboard API key through package installation, session creation, CDP connection, navigation, inspection, and cleanup.
- **429 recovery with working code.** The concurrency guide names both limits, lists four response headers, and provides retry helpers in Node.js and Python.
- **An MCP feedback loop.** The documentation MCP server identifies itself as Browserbase Documentation 1.0.0 and invites agents to submit documentation feedback.

## Start a browser session

**PASS**

Unanimous. The quickstart tells the reader where to get the API key, which packages to install, how to create the session, and how to connect with Playwright. The script has no missing variables or unexplained transition.

The page states that the API key is sufficient and the project is inferred. That removes a common first-run credential ambiguity.

## Find the exact limits

**PARTIAL**

All three judges found the same contradiction. Getting Started says the free plan includes one browser session running at a time. Plans says it includes three concurrent sessions, and its table also lists concurrency as three.

The remaining plan limits are unusually detailed. The table separates concurrency, session duration, session creation per minute, and monthly API allocations.

## Recover from a 429

**PARTIAL**

The concurrency page documents the 429 response, x-ratelimit headers, Retry-After, and exponential-backoff helpers in Node.js and Python. DeepSeek treated that as a complete recovery path.

GPT and Opus reduced the task because the Create a Session API reference lists only a 201 response. It omits the 429 that the prose says this endpoint returns. The concurrency guide also blurs the separate session-creation and concurrent-session limits in one example.

## Authenticate end to end

**PASS**

Unanimous. The docs locate the API key and optional project ID, initialize the Node.js client with the key, and show the x-bb-api-key header for direct REST requests.

The SDK reference and API reference agree that projectId is optional and can be inferred from the API key.

## Use the Node.js SDK

**PASS**

The official package is easy to find, and the minimum session example matches the quickstart and API response shape. GPT and DeepSeek passed it.

Opus found an import drift. The SDK reference uses a default Browserbase import, while the create-session guide uses a named Browserbase import. Both forms appear across the documentation.

## The receipt

> The free plan includes “One browser session running at a time” on Getting Started and “Includes 3 concurrent sessions” on Plans.

Both quotations were re-fetched from live markdown pages on 2026-08-19. All three judges reduced the limits task to PARTIAL.

- [docs.browserbase.com/welcome/getting-started](https://docs.browserbase.com/welcome/getting-started)
- [docs.browserbase.com/account/billing/plans](https://docs.browserbase.com/account/billing/plans)

## Readiness notes

The documentation MCP endpoint answered a JSON-RPC initialize POST with serverInfo Browserbase Documentation 1.0.0 and search, retrieval, resource, and feedback capabilities.

The markdown surface is complete. Every page used for the verified receipt returned authored markdown at its .md URL.

## Method note

This is a reading test of public documentation, not an execution test. No accounts were created and no API calls were run. Authentication end to end replaced webhooks because no Browserbase product-webhook interface was found during preflight. The AI Agent Readiness Score counts the fifteen reading votes at PASS 2, PARTIAL 1, FAIL 0 (30 points) and the five readiness checks at 10 points each (50 points), out of 80. Consensus tiles show each row’s majority and play no scoring role. Two of five tasks split, and every vote remains visible. Quotes shown here were re-fetched and confirmed verbatim on 2026-08-19.

Methodology: https://docsforagents.com/methodology/

Canonical URL: https://docsforagents.com/reports/browserbase-docs-ai-agent-readiness/