Kernel review, pricing and limits
Cloud Chromium browsers for AI agents, over a hosted MCP server.
- Best for
- Backend/infra engineers wiring browser-using agents into production
- Not for
- Non-technical users wanting a no-code browsing assistant
Should you use Kernel?
Kernel suits teams whose agents need real browsers with persistent logins, and it ranked first on ComputeSDK’s 6 November 2026 throughput board. The catch is trust: a security incident reported on 4 November 2026 touched 0.28% of customers, so check key scopes before connecting production agents.
Vaults, managed auth and card payments ship in the same MCP server as the browsers.
Also for
- AI product teams needing compliant automation for regulated workflows
- Developers building browser-using AI agents and computer-use loops
Show 3 more
- Teams replacing a self-hosted Playwright or Puppeteer fleet with managed infrastructure
- AI product teams needing SOC 2/HIPAA-ready browser automation for regulated industries
- Agencies automating login-gated workflows via Managed Auth and stored credentials
Main trade-off
A security incident in November 2026 exposed a small number of customer API keys and app secrets, and usage-based pricing is harder to forecast than a flat seat price.
Kernel pricing
| Plan | Monthly price | What it includes |
|---|---|---|
| Free | Free | $5 monthly usage credit, 5 concurrent browsers, live view, stealth mode, 3 managed auth connections, 3 vaults, 1-day session replays |
| Hobbyist | $30/mo | $10 monthly usage credit, 10 concurrent browsers, configurable proxies, 7-day replays, unlimited managed auth connections and vaults |
| Start-up | $200/mo | $50 monthly usage credit, 150 concurrent browsers, bring-your-own proxies, GPU acceleration, regional browsers, 30-day replays, unlimited extensions |
| Enterprise | Custom | High concurrency, 99.9% SLA, SSO, 24/7 Slack support, HIPAA BAA, zero data retention, white-labeling |
Usage above the included credit bills per second of active browser time: $0.0000166667 for a headless browser, $0.0001333336 for a headful one and $0.0008000016 for a headful browser with GPU. Idle time in Standby Mode and proxies are never charged, and a browser pool counts toward the concurrency limit even when its browsers are idle. Agents can also buy one stealth headful browser for 30 minutes at $0.50 without an account, through the machine payments protocol. Source: kernel.sh/docs/info/pricing, read 13 November 2026.
- Billing model
- Freemium
Install Kernel
brew install onkernel/tap/kernel && kernel mcp install --target claude-code
Before you start
- Install the Kernel CLI (brew install onkernel/tap/kernel), or skip it and point any MCP client at the hosted endpoint via npx -y mcp-remote https://mcp.onkernel.com/mcp
- Create a Kernel account at kernel.sh and authorize the MCP server through OAuth 2.1 or an API key
- An MCP client with remote-server support, such as Claude Code or Cursor
How to invoke it
Agents connect to the hosted MCP endpoint (https://mcp.onkernel.com/mcp) over Streamable HTTP, or through stdio via the mcp-remote bridge, and call its 33 tools (manage_browsers, computer_action, execute_playwright_code and others) by name once the server is registered in the client.
Works with
- Claude Code
- Claude Desktop
- Cursor
- VS Code (via MCP)
- Windsurf
- Zed
- Goose
- fx
- Google Antigravity
- OpenCode
- Smithery
- Playwright scripts over the CDP URL
Where it runs
- Web
- CLI
- API
Data it can reach
- Live and persistent cloud browser sessions
- Browser cookies and login state via Profiles
- Stored credentials, including TOTP codes
- Vaults, wallets and payment cards
- MP4 session replay recordings
- Kernel's own documentation index
The permissions this connector asks for, as its own documentation states them.
What Kernel can do
33 actions an agent can call. Open one for its parameters and an example.
manage_browsersCreate, update, list, get, and delete cloud browser sessions, and read archived telemetry for a session.
Example call
{"tool": "manage_browsers", "arguments": {"action": "create", "headless": false, "stealth": true, "persistence": {"id": "my-profile"}}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| action | string | Required | One of create, update, list, get, delete, get_telemetry. |
| headless | booleandefault true | Optional | Run without a visible display; headful sessions cost more but support computer_action. |
| stealth | booleandefault false | Optional | Enable anti-detection fingerprint coherence for the session. |
| profile_id | string | Optional | Attach a saved Profile so cookies and logins persist. |
| proxy_id | string | Optional | Attach a configured proxy (ISP, residential, mobile or custom) to the session. |
| session_id | string | Optional | Required for update, get, delete, and get_telemetry actions. |
Returns
| Field | Type | Description |
|---|---|---|
| session_id | string | Identifier for the created or retrieved browser session. |
| cdp_url | string | Chrome DevTools Protocol endpoint for driving the session with Playwright/Puppeteer. |
manage_profilesSet up a persistent browser profile via a guided live session, and search, list, get, or delete existing profiles.
Example call
{"tool": "manage_profiles", "arguments": {"action": "setup", "name": "github-login"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| action | string | Required | One of setup, list, search, get, delete. |
| name | string | Optional | Human-readable name for the profile. |
| profile_id | string | Optional | Required for get and delete actions. |
| query | string | Optional | Search term used with the search action. |
Returns
| Field | Type | Description |
|---|---|---|
| profile_id | string | Identifier for the saved profile, reusable across future sessions. |
manage_projectsCreate, list, get, update, and delete organization projects, and inspect their resource limits.
Example call
{"tool": "manage_projects", "arguments": {"action": "list"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| action | string | Required | One of create, list, get, update, delete. |
| project_id | string | Optional | Required for get, update, and delete actions. |
| name | string | Optional | Project display name, used with create/update. |
Returns
| Field | Type | Description |
|---|---|---|
| project | object | Project details including resource limits and scoped API keys. |
manage_api_keysCreate, list, get, update, and delete organization-wide or project-scoped API keys. The plaintext key is returned once, at creation.
Example call
{"tool": "manage_api_keys", "arguments": {"action": "create", "scope": "project", "project_id": "proj_123"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| action | string | Required | One of create, list, get, update, delete. |
| scope | stringdefault org | Optional | org or project. |
| project_id | string | Optional | Required when scope is project. |
| key_id | string | Optional | Required for get, update, and delete actions. |
Returns
| Field | Type | Description |
|---|---|---|
| plaintext_key | string | Only returned on create; not retrievable afterward. |
manage_browser_poolsCreate, list, get, delete, and flush pools of pre-warmed browsers, and acquire or release a browser from a pool.
Example call
{"tool": "manage_browser_pools", "arguments": {"action": "acquire", "pool_id": "pool_abc"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| action | string | Required | One of create, list, get, delete, flush, acquire, release. |
| pool_id | string | Optional | Required for get, delete, flush, acquire, release. |
| max_concurrent_sessions | number | Optional | Unified concurrency cap for the pool. |
Returns
| Field | Type | Description |
|---|---|---|
| browser | object | The acquired browser session, when action is acquire. |
manage_proxiesCreate, list, get, check, and delete proxy configurations across datacenter, ISP, residential, mobile, and custom proxy types.
Example call
{"tool": "manage_proxies", "arguments": {"action": "create", "type": "residential"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| action | string | Required | One of create, list, get, check, delete. |
| type | string | Optional | isp, residential, mobile, or custom. Datacenter proxies were deprecated on 16 October 2026. |
| proxy_id | string | Optional | Required for get, check, and delete actions. |
Returns
| Field | Type | Description |
|---|---|---|
| proxy_id | string | Identifier to attach to a browser session via manage_browsers. |
manage_replaysStart, stop, and list MP4 video replay recordings for a browser session. Requires a paid Kernel plan.
Example call
{"tool": "manage_replays", "arguments": {"action": "start", "session_id": "sess_123"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| action | string | Required | One of start, stop, list. |
| session_id | string | Required | The browser session to record. |
Returns
| Field | Type | Description |
|---|---|---|
| replay_url | string | URL to the MP4 recording once available. |
manage_extensionsList and delete browser extensions uploaded to the organization.
Example call
{"tool": "manage_extensions", "arguments": {"action": "list"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| action | string | Required | One of list, delete. |
| extension_id | string | Optional | Required for the delete action. |
Returns
| Field | Type | Description |
|---|---|---|
| extensions | array | List of uploaded extension metadata objects. |
Show all 33 actions
manage_appsList and search deployed apps, invoke app actions, and manage and inspect deployments and invocation results.
Example call
{"tool": "manage_apps", "arguments": {"action": "invoke", "app_id": "app_123", "action_name": "run"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| action | string | Required | One of list, search, invoke, get_deployment, list_deployments, delete_deployment, get_invocation. |
| app_id | string | Optional | Target app for invoke and deployment actions. |
| action_name | string | Optional | The specific app action to invoke. |
Returns
| Field | Type | Description |
|---|---|---|
| invocation_id | string | Identifier to poll for the invocation result. |
manage_auth_connectionsCreate, list, get, and delete managed-auth connections, and drive the login/submit/wait flow that signs an agent into a site without exposing credentials.
Example call
{"tool": "manage_auth_connections", "arguments": {"action": "login", "domain": "example.com"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| action | string | Required | One of create, list, get, delete, login, submit, wait. |
| domain | string | Optional | Domain to filter or target for the auth connection. |
| connection_id | string | Optional | Required for get, delete, submit, and wait actions. |
Returns
| Field | Type | Description |
|---|---|---|
| status | string | Connection lifecycle status, e.g. pending, authenticated, expired. |
manage_credentialsCreate, list, get, update, and delete stored credentials, and fetch current TOTP codes for a stored credential.
Example call
{"tool": "manage_credentials", "arguments": {"action": "get_totp", "credential_id": "cred_123"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| action | string | Required | One of create, list, get, update, delete, get_totp. |
| credential_id | string | Optional | Required for get, update, delete, and get_totp actions. |
Returns
| Field | Type | Description |
|---|---|---|
| totp_code | string | Current one-time code, returned only by get_totp. |
manage_credential_providersCreate, list, get, update, and delete external credential providers, such as a connected 1Password vault.
Example call
{"tool": "manage_credential_providers", "arguments": {"action": "list"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| action | string | Required | One of create, list, get, update, delete. |
| provider_id | string | Optional | Required for get, update, and delete actions. |
Returns
| Field | Type | Description |
|---|---|---|
| provider | object | Provider connection details, excluding raw secrets. |
get_connection_contextInspect the authenticated principal, organization, credential scope, and effective project scope for the current MCP session.
Example call
{"tool": "get_connection_context", "arguments": {}}Returns
| Field | Type | Description |
|---|---|---|
| organization_id | string | The authenticated organization. |
| scope | string | Effective credential scope, e.g. org or a specific project. |
computer_actionSend mouse, keyboard, clipboard and screenshot controls to a live browser session.
Example call
{"tool": "computer_action", "arguments": {"session_id": "sess_123", "operation": "click", "x": 420, "y": 180}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| session_id | string | Required | The browser session to control. |
| operation | string | Required | One of click, type, press_key, scroll, move, get_position, read_clipboard, write_clipboard, screenshot. |
| x | number | Optional | X coordinate for click, move, or scroll. |
| y | number | Optional | Y coordinate for click, move, or scroll. |
| text | string | Optional | Text to type or write to clipboard. |
Returns
| Field | Type | Description |
|---|---|---|
| screenshot_url | string | Returned when operation is screenshot. |
browser_curlSend an HTTP request through a live browser session's own Chrome network stack, preserving its cookies, proxy, and fingerprint.
Example call
{"tool": "browser_curl", "arguments": {"session_id": "sess_123", "url": "https://example.com/api", "method": "GET"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| session_id | string | Required | The browser session whose network stack issues the request. |
| url | string | Required | Target URL. |
| method | stringdefault GET | Optional | HTTP method. |
Returns
| Field | Type | Description |
|---|---|---|
| body | string | Raw response body. |
| status | number | HTTP status code. |
execute_playwright_codeRun Playwright/TypeScript code against an existing browser session and return the result.
Example call
{"tool": "execute_playwright_code", "arguments": {"session_id": "sess_123", "code": "return await page.title();"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| session_id | string | Required | The browser session to execute against. |
| code | string | Required | Playwright/TypeScript code, executed with `page` in scope. |
Returns
| Field | Type | Description |
|---|---|---|
| success | boolean | Whether the code ran without error. |
| result | string | The returned value, e.g. "Example Domain" for a page.title() call. |
exec_commandRun a shell command inside the browser session's VM and return decoded stdout/stderr.
Example call
{"tool": "exec_command", "arguments": {"session_id": "sess_123", "command": "ls /tmp"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| session_id | string | Required | The browser session's VM to run the command in. |
| command | string | Required | Shell command to execute. |
Returns
| Field | Type | Description |
|---|---|---|
| stdout | string | Decoded standard output. |
| stderr | string | Decoded standard error. |
search_docsSearch Kernel's own documentation from within an MCP client.
Example call
{"tool": "search_docs", "arguments": {"query": "standby mode pricing"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| query | string | Required | Search terms. |
Returns
| Field | Type | Description |
|---|---|---|
| results | array | Matching documentation pages with excerpts. |
open_auth_loginOpen a secure interactive Managed Auth MCP App after user consent, for clients that declare MCP Apps support.
Example call
{"tool": "open_auth_login", "arguments": {"connection_id": "conn_123"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| connection_id | string | Required | The managed-auth connection to open the interactive login for. |
manage_browser_filesRead, write, upload, download and manage files in a running browser VM; read and download refuse files over max_bytes instead of truncating them.
Example call
{"tool": "manage_browser_files", "arguments": {"action": "read", "session_id": "sess_123", "path": "/tmp/report.txt"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| action | string | Required | One of list, get_info, read, download, write, upload, upload_zip, download_dir_zip, create_directory, move, delete_file, delete_directory, set_permissions. |
| session_id | string | Required | Browser session id or name. |
| path | string | Optional | Absolute file or directory path in the browser VM; required by most actions. |
| max_bytes | numberdefault 10485760 | Optional | Maximum bytes to return for read, download and download_dir_zip; up to 25 MiB. |
Returns
| Field | Type | Description |
|---|---|---|
| content | string | File contents for read; binary downloads come back as embedded MCP resources. |
manage_config_registryLook up recommended browser and proxy settings for a target site, start and inspect analyses, and list project configurations or analysis history.
Example call
{"tool": "manage_config_registry", "arguments": {"action": "lookup", "url": "https://example.com"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| action | string | Required | One of lookup, resolve, get_analysis, cancel_analysis, list_configs, list_analyses. |
| url | string | Optional | Public http(s) target URL, for lookup and resolve. |
| intent | string | Optional | Plain-language workload to exercise during analysis, for resolve; https targets only. |
| analysis_id | string | Optional | Analysis id returned by resolve or the list actions, for get_analysis and cancel_analysis. |
Returns
| Field | Type | Description |
|---|---|---|
| recommendation | object | Recommended browser and proxy configuration for the target. |
manage_playwright_executorsList and delete the Playwright executors of a browser session; a browser can have at most 8 named executors.
Example call
{"tool": "manage_playwright_executors", "arguments": {"action": "list", "session_id": "sess_123"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| action | string | Required | One of list, delete. |
| session_id | string | Required | Browser session id or name. |
| name | string | Optional | Executor name, required for delete; deleting default restarts it. |
| close_tab | booleandefault true | Optional | Whether delete closes the executor tab. |
Returns
| Field | Type | Description |
|---|---|---|
| executors | array | Every executor with busy, created_at, last_used_at and, for named executors, target_id and url. |
web_searchSearch the web, retrieve retained results and fetch page contents; creating a search is billable and appears only for accounts with a Search entitlement.
Example call
{"tool": "web_search", "arguments": {"action": "providers"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| action | string | Required | One of create, get, contents, providers. |
| request | object | Optional | Search request, required for create. |
| search_id | string | Optional | Retained search id, required for get and contents. |
| slug | string | Optional | Optional provider filter for providers. |
Returns
| Field | Type | Description |
|---|---|---|
| search_id | string | Identifier of the retained search, returned by create. |
browser_replRun JavaScript in a persistent Node.js runtime inside a browser VM, with native browser helpers, WebMCP, Patchright, Playwright and raw CDP; this is unrestricted VM code execution, not a sandbox.
Example call
{"tool": "browser_repl", "arguments": {"session_id": "sess_123", "code": "return await page.title();"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| session_id | string | Required | Browser session id or name. |
| code | string | Optional | JavaScript for the cell; top-level bindings survive across calls. |
| reset | boolean | Optional | Restart the runtime before running the cell. |
| timeout_sec | number | Optional | Per-call timeout in seconds. |
Returns
| Field | Type | Description |
|---|---|---|
| result | string | The cell output. |
webmcpDiscover tools that pages expose across tabs and frames, invoke one by its exact tool_ref, and manage custom tool definitions.
Example call
{"tool": "webmcp", "arguments": {"action": "list", "session_id": "sess_123"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| action | string | Required | One of list, invoke, list_custom, add_custom, remove_custom. |
| session_id | string | Required | Live browser id or name. |
| tool_ref | string | Optional | Reference from the latest list, required for invoke. |
| input | object | Optional | Input matching the tool's inputSchema, for invoke. |
| timeout_sec | numberdefault 60 | Optional | Invoke timeout, 1 to 120 seconds. |
Returns
| Field | Type | Description |
|---|---|---|
| status | string | completed, canceled, error or awaiting_submission, with an invocation_id. |
manage_vaultsCreate, list, get and delete project-owned vaults; use one per end user.
Example call
{"tool": "manage_vaults", "arguments": {"action": "list"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| action | string | Required | One of create, list, get, delete. |
Returns
| Field | Type | Description |
|---|---|---|
| vault | object | The vault record. |
manage_vault_provider_configsCreate, list, get, rename, rotate secrets for and delete organization-owned Link and AgentCard configurations; writes need organization scope.
Example call
{"tool": "manage_vault_provider_configs", "arguments": {"action": "list"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| action | string | Required | One of create, list, get, update, delete. |
Returns
| Field | Type | Description |
|---|---|---|
| config | object | The provider configuration, without raw secrets. |
manage_vault_walletsConnect Kernel-hosted card wallets or Kernel-managed Link and AgentCard wallets and inspect live payment methods and eligibility.
Example call
{"tool": "manage_vault_wallets", "arguments": {"action": "payment_methods"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| action | string | Required | One of create, payment_methods. |
Returns
| Field | Type | Description |
|---|---|---|
| wallet | object | Wallet details and eligibility. |
manage_vault_cardsCreate or update card requests and authorize an advertised card after user approval; creating or updating a card never authorizes it.
Example call
{"tool": "manage_vault_cards", "arguments": {"action": "create"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| action | string | Required | One of create, update, authorize. |
Returns
| Field | Type | Description |
|---|---|---|
| card | object | Card request state, including any hosted spend-approval action. |
manage_vault_credentialsCreate credentials through Kernel-hosted collection or through a 1Password approval flow, or connect a provider account.
Example call
{"tool": "manage_vault_credentials", "arguments": {"action": "create"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| action | string | Required | One of create, update, connect_account. |
Returns
| Field | Type | Description |
|---|---|---|
| credential | object | Credential definition; sensitive values stay hidden. |
manage_vault_itemsList and get vault items, invoke advertised operations such as fill, webmcp_invoke and confirm_transaction, observe events and delete items.
Example call
{"tool": "manage_vault_items", "arguments": {"action": "list"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| action | string | Required | One of list, get, invoke, events, delete. |
Returns
| Field | Type | Description |
|---|---|---|
| status | string | Operation status; ready is not login or payment success. |
get_more_toolsReport a missing KERNEL capability, integration or site-specific WebMCP action after checking the available tools.
Example call
{"tool": "get_more_tools", "arguments": {"capability_area": "webmcp"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| capability_area | string | Required | The area of the gap, for example webmcp. |
| task_outcome | string | Optional | Normalized outcome of the task that hit the gap. |
submit_feedbackSend product, site-compatibility, config-registry, MCP or documentation feedback to the KERNEL team without interrupting the current task.
Example call
{"tool": "submit_feedback", "arguments": {"summary": "Docs example for manage_profiles is out of date", "feedback_type": "documentation"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| summary | string | Required | Short statement of the feedback. |
| feedback_type | string | Required | The kind of feedback being reported. |
| task_outcome | string | Optional | Normalized outcome of the task. |
What Kernel is for
Kernel is a hosted Model Context Protocol server that lets Claude Code, Cursor and other MCP clients drive cloud Chromium browsers. Its README lists 33 tools covering sessions, profiles, proxies, vaults and Playwright execution, and it connects over Streamable HTTP with OAuth 2.1 authentication.
Fast cold starts, vendor-reported
Kernel's homepage says it spins up browsers in under 30 ms and boots from a cold request to a ready browser in 794 ms, against 2,967 ms for Browserbase in the steel-dev browserbench table it cites.
Top of an independent throughput board
Kernel ranked first of seven providers on ComputeSDK's public browser throughput board in its 6 November 2026 run, which times ten sequential actions across 100 stealth sessions per provider.
Standby Mode with zero idle cost
Idle browsers suspend and resume on the next request, and idle browsers in a pool incur no usage charges, so only active time is billed.
Stealth proxies and CAPTCHA solving
Free plans include stealth mode with a Kernel-managed residential proxy and CAPTCHA auto-solver, and residential traffic now fails over across several providers.
Show all 8 features
Profiles, Managed Auth and vaults
Saved profiles persist cookies and logins, Managed Auth does the login for the agent and keeps the credential away from it, with no per-connection fee, and vaults store credentials per end user, up to 3 vaults on the free plan.
Agentic payments
Agents can pay at checkout through Link by Stripe wallets or a user-approved Visa card, with no surcharge for the payments products.
Session observability
Live view, MP4 replays kept for 1 day on Free, 7 on Hobbyist and 30 on Start-up, and a telemetry API let an agent debug its own failed runs.
Compliance posture
SOC 2 Type II applies to every plan, ISO 27001 was announced in June 2026, HIPAA with a BAA is on Enterprise, and GDPR is listed as in progress.
What people use it for
Give an MCP agent hands on the live web
Register the hosted MCP endpoint in Claude Code or Cursor so the agent can call manage_browsers and computer_action directly to click, type, and read real pages instead of relying on stale training data.
Run CAPTCHA-gated scraping jobs unattended
Route requests through Kernel's residential and ISP proxies with automatic CAPTCHA solving so a scraping agent keeps working on sites that would otherwise block headless traffic.
Persist a logged-in session across agent runs
Create a browser Profile once so cookies and auth survive between invocations, letting a scheduled agent skip re-login on every run.
Debug a failed automation from a video replay
Pull the MP4 replay for a session via manage_replays to see exactly where a Playwright script or computer-use loop diverged from the expected page state.
Let an agent pay at checkout with a card the user approves
Create a vault wallet and a card request, have the user approve it, then let the agent fill and confirm the purchase through manage_vault_items.
Strengths and limits
Strengths
- Kernel's pricing docs, read on 13 November 2026, say idle time is not charged: Standby Mode and idle pool browsers cost nothing until a session is active.
- The server's README lists browser sessions, Playwright execution, shell commands, vaults and payments as tools of one MCP server, so an agent needs no separate integration for each.
- Median of 7.58 actions per second on the latest ComputeSDK board, against 5.32 for Browserbase and 4.03 for Steel.
- Its plan table lists SOC 2 compliance on every plan and a HIPAA BAA on Enterprise only.
Limits
- Kernel disclosed a security incident on 4 November 2026 that exposed a small number of API keys and app secrets and touched 0.28% of customers; scoped API keys are its main remedy, so audit key scopes.
- Free Claude plans cannot add the remote MCP server, because it does not work with their method of adding servers.
- Headful sessions with GPU cost 48 times a headless session per second (about $0.0008 against $0.0000167), so GPU-heavy workloads need active monitoring.
- Pricing is usage-based on top of the plan fee, which makes cost harder to predict for spiky, high-concurrency workloads than a flat-fee product.
Show 2 more
- Session replays on the free plan are kept for 1 day, and longer retention needs Hobbyist or Start-up.
- Stealth mode and CAPTCHA solving can conflict with a site's terms of service, so check them before automating a site.
Kernel alternatives
The nearest AI agent skills on Hokai for the same job.
| Skill | Works with | Connects by | Checked |
|---|---|---|---|
| Kernel This page | Claude · Cursor · Windsurf · VS Code | MCP | 13 Nov |
 MantisA local browser for agents that returns compact numbered text, not screenshots. | Claude · Codex · Cursor | MCP | 11 Nov |
 NotteCloud browsers, browser agents and deployable browser functions for AI, with a CLI and two MCP servers. | Claude · Cursor · Windsurf | MCP | 11 Oct |
 OlostepScrape, crawl, map, search and answer from the web, as an API, a CLI and an MCP server. | Claude · Cursor · Windsurf · VS Code | MCP | 11 Oct |
Kernel facts
- Protocol
- MCP
- Authentication
- OAUTH
- Latency class
- Fast
- Open source
- Yes
- Maintainer founded
- 2025
- Headquarters
- San Francisco, CA, USA
- Team size
- 22
- Rate limits
- Concurrent browsers are capped by plan: 5 on Free, 10 on Hobbyist and 150 on Start-up (kernel.sh/pricing, 13 November 2026); Enterprise concurrency is custom.
- Total funding
- Seed and Series A, $22M in total (October 2025), led by Accel with Y Combinator, Cintrifuse Capital, Vercel Ventures, Refinery Ventures and SV Angel
What changed
Questions about Kernel
How much does Kernel cost?
Kernel has a free plan, Hobbyist at $30 a month, Start-up at $200 a month and a custom Enterprise plan, with included usage credit of $5, $10 and $50 on the first three. Credit covers usage only. Beyond it a headless browser costs about $0.06 an hour and a headful one about $0.48, billed for active use, and idle time and proxies are not charged. Source: kernel.sh/pricing, read 13 November 2026.
Is Kernel open source, and can you try it free?
The Kernel MCP server is MIT-licensed and the kernel-images browser build is Apache-2.0, both on GitHub. The hosted service has a free plan with 5 concurrent browsers, live view and stealth mode, and signing up needs no credit card. Paid plans add concurrency, longer replay retention and configurable proxies.
Which agents and clients can connect to Kernel's MCP server?
Kernel documents setup for ten clients, among them Claude, Cursor, VS Code, Windsurf and Zed. Any other client with remote MCP support can use the Streamable HTTP endpoint, and a client without it can fall back to mcp-remote over stdio. The one known exclusion is free Claude plans, whose way of adding servers does not work with Kernel.
Has Kernel had a security incident?
Yes. Its notice of 4 November 2026 describes an incident detected on 2 November, in which a leaked API key gave an attacker the data held at a logging provider and then four GPU servers, three of which ran customer browser sessions. The company puts the affected share at 0.28% of customers, says it revoked and rotated the credentials involved, and is promoting scoped API keys.
How does Kernel compare with Browserbase on speed?
In ComputeSDK's public benchmark, last run on 6 November 2026, Kernel scored 87.9 against 77.9 for Browserbase, at 7.58 and 5.32 actions per second and a median task of 1.32 and 1.88 seconds. Its test runs ten actions in a row on Wikipedia, 100 sessions per provider, so time your own login flow before you commit.
About Kernel
Kernel is a hosted MCP server that gives an agent cloud Chromium browsers to drive: it can create a session, click and type in it, run Playwright code against it, keep logins in saved profiles and replay what happened as video. It comes from Kernel, a San Francisco company founded in 2025 by Catherine Jue (CEO, previously co-founder of Sway Finance) and Rafael Garcia (CTO). The company went through Y Combinator's Summer 2025 batch and announced a $22 million seed and Series A round led by Accel in October 2025. Y Combinator's company page lists 22 people, and Kernel's homepage claims 11,000+ teams build on it. Its users are agents, not people: they invoke it through MCP or the API, and the dashboard exists for keys, billing and watching sessions.
Read the full background
The server runs at mcp.onkernel.com over Streamable HTTP and authenticates with OAuth 2.1 and dynamic client registration; clients without remote MCP support can use the mcp-remote stdio bridge. Its README lists 33 tools. Each Kernel feature has one manage_* tool with an action parameter (browsers, pools, profiles, proxies, vaults, replays, apps and so on), beside standalone tools for mouse and keyboard control, Playwright execution, a persistent JavaScript REPL and WebMCP page tools. Some tools only appear for clients or accounts that qualify, such as web search and the interactive login app.
Any client that speaks remote MCP can connect. Kernel's docs have setup pages for Claude, Cursor, Google Antigravity, OpenCode, fx, Goose, VS Code, Windsurf, Zed and Smithery, and the Kernel CLI wires several of them up in one command. It fits Claude Code and Cursor sessions that need a real browser, scheduled agents that must stay logged in, and checkout flows where a user approves a card. The best MCP servers guide ranks the browsing options by job, and the browser skills directory lists the rest.
Where an alternative wins: Browserbase posted a faster median screenshot, 142 ms against Kernel's 189 ms, in ComputeSDK's public run of 6 November 2026, where Kernel led on actions per second and on composite score. ComputeSDK's test is ten sequential actions on Wikipedia across 100 sessions with stealth on, so it says little about login-heavy flows, and Browser Use had a shorter median task time, 0.11 seconds against 1.32. ComputeSDK's board also lists Hyperbrowser, Steel and Browser Use as hosted rivals, with composite scores of 26.7, 54.4 and 34.9, and Kernel's homepage compares cold starts with Anchor Browser. Free Claude users cannot add Kernel at all: its README says the remote server does not work with the method free Claude plans use to add MCP servers.
What shipped recently: on 6 October 2026 Kernel added named Playwright executors, so several tasks can run in parallel tabs of one browser. On 16 October it shipped scoped API keys, general availability of Config Registry on Hobbyist plans and up, Visa card payments for agents through the CLI and MCP server, a larger GPU tier and failover across several residential proxy providers, and it deprecated datacenter proxies. It announced ISO 27001 certification in June 2026. On 4 November 2026 it published a notice of a security incident: an attacker used a compromised API key from 24 October, reached a logging provider and four GPU servers, and Kernel says 0.28% of its customers were affected.
Who should pick what: choose Kernel if your agents run long sequences of actions or need vaults and card payments beside the browser, since it posted 7.58 actions per second to Browserbase's 5.32. Choose Browserbase if your loop waits mostly on screenshots, where its 142 ms median beat Kernel's 189 ms. Whichever you pick, give each agent a scoped API key.
Sources and guides
Official sources
- Kernel official site Vendor
- GitHub repository GitHub
- Community Discord
Protocol, authentication, permissions and the call contracts on this page are read from Kernel's own documentation. The verdict is Hokai's reading of them, not the maintainer's claim.