comparison

browser-memory vs Playwright MCP

Playwright MCP hands your agent raw browser control. browser-memory hands it pre-built, per-site skills. One is how to move the browser, the other is what to do on a given site, and they work together.

browser-memoryUpdated Jul 2026
browser-memoryPlaywright MCP
What it gives the agentPre-built per-site skillsRaw browser control primitives
LevelKnowledge of a specific siteClick, type, navigate, snapshot
Per-run workCall the skill, get dataDerive every site action live
Output to the modelStructured fieldsAccessibility snapshot of the page
When a site changesSkill repaired once, for all agentsAgent re-derives the steps itself
RelationshipSits on top of a browserDrives the browser
Installbmem add {site}/{task}Add the MCP server to your agent

Control vs knowledge

Playwright MCP is an excellent primitive: it lets an agent click, type, navigate and read a page snapshot through a clean MCP interface. But it is control, not skills. Every site-specific action, finding the search box, reading a profile, building a cart, still has to be worked out live, on every run, from the raw page.

browser-memory sits one layer up. Each repeated site action is already a pre-built tool made of the exact selectors and requests, returning structured fields. The agent calls it instead of re-deriving it, which is where the tokens and the wall-clock go.

Better together

This is not an either/or. Both speak MCP, so an agent can hold raw browser control and browser-memory skills at once: primitives for the long tail, pre-built skills for everything the ecosystem has already learned. When a skill breaks, it is repaired once for every agent, so you are not maintaining page logic by hand.

FAQ

What is the difference between browser-memory and Playwright MCP?

Playwright MCP is a Model Context Protocol server that exposes raw browser control to an agent: navigate, click, type, and read a page snapshot. browser-memory is a layer of pre-built, per-site skills on top of a browser. Playwright MCP gives control; browser-memory gives knowledge of how to do a specific task on a specific site, returning structured data.

Can I use them together?

Yes, they are complementary. Both speak MCP. You can drive a browser with control primitives and still call browser-memory skills for the site-specific actions, so the agent does not have to re-derive selectors and requests every run.

Why not just use control primitives for everything?

Because with only primitives the agent re-reads and re-figures out each page every run, which is slow and token-heavy. Pre-built skills turn a repeated site action into a single tool call that returns fields, which is where most of the latency and token savings come from.

Is browser-memory MCP-compatible?

Yes. browser-memory exposes its skills over MCP, so any MCP-capable agent, including Claude Code, can discover and run them.

Compare more: browser-use, browse.sh, Browserless, Playwright, Stagehand, Browserbase, Unbrowse.

Give control something to call

Keep your browser primitives. Add pre-built, self-repairing skills so your agent stops re-deriving every page.