Ostler MCP browser documentation
Ostler is an open-source MCP browser for AI coding agents such as Claude Code, Codex and Cursor. It reads, researches and acts in a real Chrome, and checks its own work: research answers cite only pages it actually read, with every quote verified; forms are filled in one call with every value read back; and Qdrant remembers what worked. It speaks MCP to coding agents and A2A to other agents.
Get started
Get started without programmingInstall Ostler with a double-click, an editor extension, pip or uvx, or one command: no Rust and no config files.
QuickstartInstall, check, connect, then read and cite a demo page or fill a demo form, in five minutes.
InstallEvery way to get the binary, what works today, update and uninstall.
Use with your agentAdd Ostler as an MCP browser to Claude Code, Codex, Cursor, VS Code or any MCP client, with a forms and a research workflow.
LangChain, LlamaIndex, AutoGenUse Ostler's tools from three Python agent frameworks: tested code, real output, and the one-session rule for cite.
Guides
Research with checked citationsRead pages, check each quote against them, and report what could not be read, with real tool output.
CookbookRecipes: deep research with checked citations, reading, QA, logins, forms, approvals and A2A, with what is not built yet.
Forms and verificationHow keys become fields, how values are verified by read-back, and what submit results mean.
Memory and privacyWhat is stored, what sealing hides and does not, local embeddings, per-tenant tokens.
Configuration and CLIConfig file, environment variables, every serve and a2a flag.
Reference
Tools referenceAll 32 MCP tools: browse, read and cite, search, extract, forms and memory, with parameters and profiles.
Security modelThe threat model: what Ostler does, and what it does not do.
A2ALet other agents hand Ostler forms over Agent2Agent v1.0.
ArchitectureCrates, the injected runtime, the form model, and the decisions behind them.
Evidence
BenchmarksMeasured on local fixtures with a scripted agent. Read the caveats.
Ostler vs Playwright MCPFive scripted form tasks on local fixtures: tool calls, tokens, time, and what the comparison does not show.
EconomicsUnit costs, scaling limits, the moat question and what could go wrong.
Design reviewThe October 2026 review: five design flaws and their status.