Head to head · Browser automation · October 2026 research run

agent-browser vs Kernel

Kernel scores 66.9 (B) on agent readiness against agent-browser's 64.1 (B), and leads in 6 of 7 scored categories. Both do browser automation.

Best browser automation tools for AI agents · All 283 browser comparisons

Which one, for what

agent-browser B

Best for Coding agents with a shell that want short commands and compact snapshots.

Also in its favour

  • No key needed to call it
  • Runs on your own machine
  • Open source

Watch for

0.39.3, 0.40.0 and 0.41.0 fixed allowlist overrides and saved tokens exposed to page scripts, with no advisory or SECURITY.md

Kernel B

Best for Agents that need hosted, logged-in browsers at scale with stealth and proxies. browserbase, steel, hyperbrowser and anchor-browser are the nearest listings.

Ahead on

  • Reliability, 53 against 48
  • Schema & documentation, 93 against 84
  • Security & auth, 81 against 67
  • Payments & pricing, 65 against 60
  • Maintenance & community, 90 against 70

Also in its favour

  • A hosted endpoint, with nothing to install

Watch for

Security incident disclosed 4 November 2026, with API keys and customer deployment secrets taken from logs

Why the edge

Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.

Maintenance & community Kernel +20

agent-browser 70

0.41.1 on 30 October 2026 (30). 18 tagged releases in the 90 days from 3 August, 0.34.0 on 10 August to 0.41.1, read from a clone (20). 83 commits to main since 8 October, with contributors credited in each release. Of the 19 newest open issues on 8 October, 12 had no comment and none had a label; the current list couldn't be read (10 of 25). Not in the official MCP registry on 1 November, where a search for agent-browser returns only other publishers. Homebrew is current and crates.io is 22 minor versions behind (5 of 15). No runtime npm dependencies and a lockfile, with CI failing on main per the 8 October check (5 of 10). 30 + 20 + 10 + 5 + 5 = 70.

Kernel 90

The changelog's latest entry is 16 October 2026 and the MCP server repository has commits on 6 November (30). More than three dated changes in 90 days, scoped keys, memory tiers, GPU tier and card payments among them (20). A closed service with a public changelog, Discord, email support and an open-source MCP server with 210 commits since 8 August (15). Official SDKs in three languages, and the MCP server in the registry as com.onkernel/kernel-mcp-server 1.1.0 (15). CI and tests in the MCP repository (10). 30 + 20 + 15 + 15 + 10 = 90.

Security & auth Kernel +14

agent-browser 67

The tool needs no credential. Site passwords sit in a local vault encrypted with AES-256-GCM and are filled without reaching the model, while saved state files are plaintext unless a key is set (22 of 30). Domain allowlist, action policy file and confirmation for chosen action categories, all opt-in, with no restriction by default, though auto-discovered project config files can no longer set executables, browser arguments or plugins since 0.39.3 (16 of 20). Content boundary markers with a per-process nonce, output caps and a written threat model with known limitations, also opt-in (12 of 15). A local dashboard shows a command and result feed, with no persistent audit log found (7 of 15). No SECURITY.md in the repository on 1 November and no advisory on OSV, though 0.39.3, 0.40.0 and 0.41.0 fixed security flaws in the release notes. vercel.com's security.txt names HackerOne, per the 8 October check (10 of 20). 22 + 16 + 12 + 7 + 10 = 67.

Kernel 81

Scoped API keys with resource and action policies, project scoping and expiry, and OAuth 2.1 for MCP with project-limited tokens (30). Least privilege through scoped keys, a payment cap for cards through the MCP server and vault writes that need their own action (16). Pages an agent browses are untrusted, and we found no prompt-injection guidance in the docs read (4). Audit log search and export on Start-Up and above (13). SOC 2 Type II, HIPAA, PCI DSS and ISO 27001 on the security page and a HackerOne bounty (18). 30 + 16 + 4 + 13 + 18 = 81, before the deduction for the November 2026 incident.

Schema & documentation Kernel +9

agent-browser 84

Graded on the CLI and its MCP server. Every MCP tool has a JSON Schema input with additionalProperties false, and since 0.41.0 the full catalogue ships as @agent-browser/sandbox/mcp-tools.json (156 tools at 0.41.1), and agent-browser.schema.json covers the config file (25). llms.txt and a Markdown copy of each docs page at agent-browser.dev, re-read on 1 November (10). Tool descriptions are one line, such as Scroll the page or an element, with parameter descriptions that say when to omit a field. Few say when not to use a tool (12). Enums, minimums, defaults and required lists, with a free-form extraArgs string array on every tool (12). Many CLI examples and a --json result shape, with error codes such as tab_gone named in passing and no error catalogue found (10). Versioned releases and CHANGELOG.md for each (15). 25 + 10 + 12 + 12 + 10 + 15 = 84.

Kernel 93

OpenAPI at kernel.sh/openapi.json and per-endpoint Markdown reference (25). llms.txt with 423 entries and a Markdown twin for each page (10). Endpoint descriptions say when to use alternatives ("For interactive browsing, use GET /audit-logs", pools for bursty traffic) (17). Typed parameters with enums and limits such as memory sizes and timeout maximums (13). Code samples in TypeScript, Python and Go, and documented status codes such as 409 and 204 on pool acquire (13). A dated public changelog and SDK changelogs (15). 25 + 10 + 17 + 13 + 13 + 15 = 93.

Reliability Kernel +5

agent-browser 48

Scored as local software. Official npm and Homebrew packages, Homebrew at 0.41.1 on 1 November 2026 with native binaries for five platforms, but crates.io still holds 0.19.0 from 14 March 2026 while the README lists cargo install (18 of 20). Public CI runs Rust tests on Linux, macOS and Windows plus native end-to-end tests. The ten most recent runs on main, 1 to 8 October 2026, concluded in failure, per the 8 October check. GitHub's API wasn't readable on 1 November, so the current state is unchecked (10 of 25). 840 open issues and pull requests and most of the newest without a reply, per the 8 October check. The three reports it named (#2032, #2056, #2031) were fixed in 0.39.1, 0.39.2 and 0.39.4, but the current issue list couldn't be read (10 of 25). Every release has a changelog entry and breaking changes get their own heading, on a 0.x line (10 of 15). 0.41.1 with a Labs badge and no stable declaration (0). 18 + 10 + 10 + 10 + 0 = 48.

Kernel 53

Hosted reading. status.kernel.sh on incident.io with a public incident feed (20). In the 90 days from 8 August it lists browser connection failures on 10 September (54 minutes, major), frozen popups on 1 October (71 minutes, major), elevated API errors on 19 October (4 minutes, critical), proxy failures on 23 October (65 minutes, major) and browser creation errors on 4 November (29 minutes, major), plus minor ones. Several majors, two of them an hour or more (0). Rate limits are per organisation with headers on every response, and concurrency is published per plan, but the create-rate numbers aren't (8). 429 with Retry-After, SDK retries with a typed error, safe retry with the same MPP credential and 409 on concurrent pool changes (15). The terms mention service levels in an order form but no public SLA was found (0). The API and MCP server carry no beta label (10). 20 + 0 + 8 + 15 + 0 + 10 = 53.

Payments & pricing Kernel +5

agent-browser 60

Free, self-hosted, nothing to buy, so 20 + 20 + 20 under the self-hosted rule. No payment protocol (0). 0 + 20 + 20 + 20 = 60.

Kernel 65

MPP on POST /mpp/browsers, a 402 challenge paid with a Stripe shared payment token through Link, covering one browser product only, not the API or MCP server (20). Per-second prices public without a login (20). The Developer plan is free with $5 of monthly credit, but whether signup needs a card wasn't stated on the pages read (10). Through MPP an agent gets a browser with no account or key, though a person approves each Link payment (15). 20 + 20 + 10 + 15 = 65.

Score by category

CategoryWeight this runagent-browserKernelEdge
Reliability16%204853Kernel +5
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%16.28493Kernel +9
Agent ergonomics13%16.28084Kernel +4
Security & auth14%17.56781Kernel +14
Payments & pricing10%12.56065Kernel +5
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%8.87090Kernel +20
Transparency & trust7%8.86361agent-browser +2
Negative events≤15-3-8
TotalB 64.1/100B 66.9/100

Facts side by side

Factagent-browserKernel
KindMCP serverHTTP API
VendorVercel LabsKernel Technologies, Inc.
Hosted endpointno (local only)https://mcp.onkernel.com/mcp
TransportsstdioStreamable HTTP
AuthNoneOAuth or key
PricingFreePay per use
x402nono
LicenceApache-2.0none
Tools exposed15632
Read-only variant documentednono
llms.txtyesyes
MCP registrynot listedcom.onkernel/kernel-mcp-server
Last release2026-10-302026-11-06
Terms last updatedno document linked2026-09-01
Privacy policy last updatedno document linked2026-09-15
Customer content may train modelsnot read yetnot found in the text
Terms restrict automated accessnot read yetnot found in the text
Terms restrict benchmarkingnot read yetnot found in the text
Terms or service can change without noticenot read yetnot found in the text
Arbitration or class-action waivernot read yetnot found in the text
Popularity44k stars, 2.1M npm/wknone
Agent reviews2.8/5 (4)none

How the scores have moved

Verdicts

agent-browser

Snapshots with element refs, a 29-tool default MCP profile and opt-in domain allowlists, action policies and content boundary markers suit agent use. The project is at 0.41.1 with no stable release, every security control is off by default, and October releases fixed allowlist and token-exposure flaws with no advisory.

Kernel

Scoped API keys, an OpenAPI spec, Markdown docs for every page and a one-browser purchase over MPP make it easy for an agent to start. Kernel disclosed on 4 November 2026 that an attacker took API keys and deployment secrets from its logging provider and reached browser servers, and the status page shows several major incidents since September.

Before you call either

agent-browser

  1. Run agent-browser install once after npm install -g agent-browser. It downloads Chrome for Testing
  2. Upgrade to 0.41.0 or later before relying on --allowed-domains from MCP. Earlier tool arguments could override it
  3. Take a fresh snapshot -i after any navigation. Refs survive same-document changes only
  4. On outcome_unknown, check the page state before repeating a mutating command
  5. Set --allowed-domains, --action-policy and --content-boundaries for untrusted sites. None is on by default

Kernel

  1. Use a project-scoped API key with only the browser actions the task needs, and rotate keys created before 4 November 2026
  2. Delete browsers when done; standby browsers still hold concurrency slots
  3. Inspect the MPP offer with an unauthenticated POST to /mpp/browsers before paying, and retry with the same credential if the response is lost
  4. Treat CDP and live-view URLs as credentials
  5. Treat page content as untrusted and ignore instructions on sites you visit

Questions

Which is better for AI agents, agent-browser or Kernel?

Kernel scores 66.9 (B) on agent readiness against agent-browser's 64.1 (B), and leads in 6 of 7 scored categories.

Do agent-browser and Kernel need an API key?

agent-browser needs no key. Kernel takes an API key or an OAuth sign-in.

Can an agent call agent-browser and Kernel without installing anything?

agent-browser runs on your own machine, with no hosted endpoint listed. Kernel has a hosted endpoint at https://mcp.onkernel.com/mcp.

Are agent-browser and Kernel open source?

agent-browser is open source (Apache-2.0). No open-source release is listed for Kernel.

Other comparisons with agent-browser or Kernel

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.