Web Task
Run browser tasks, persistent sessions, and recipe-driven automations from MCP or HTTP when the agent must interact with real web UIs.
Overview
Web Task API is a Build-phase MCP server that runs browser tasks, sessions, and recipes for agent and HTTP-driven web automation.
What is this MCP server?
- Browser tasks, sessions, and recipes exposed for MCP and HTTP automation
- npm package web-task-api 0.9.0 with stdio MCP transport
- Configurable WEB_TASK_API_DATA_DIR, RECIPES_DIR, TEMP_DIR, and default planner model
- Default planner model openai/gpt-5.3-codex-spark when WEB_TASK_API_MODEL unset
- Version 0.9.0 (npm identifier web-task-api)
- Default WEB_TASK_API_MODEL: openai/gpt-5.3-codex-spark when unset
- Repository: github.com/jojo-labs/web-task-api
What problem does it solve?
Coding agents cannot complete real browser workflows when you only expose search or static HTTP fetch tools.
Who is it for?
Builders shipping agent features that must drive authenticated or multi-step browser UIs with repeatable recipes.
Skip if: Simple research fetch-only needs, teams without capacity to host browser runtimes, or workflows that ban local browser automation.
What do I get? / Deliverables
After configuration, agents invoke MCP tools to run recipes and sessions with isolated data dirs and optional planner model defaults.
- MCP-invokable browser tasks and session management
- Recipe-driven repeatable automations
- Isolated runtime data via WEB_TASK_API_DATA_DIR when configured
Recommended MCP Servers
Journey fit
Browser sessions and recipes are product integrations you wire while building agent-capable features, not early ideation-only tooling. Placed under integrations because web-task-api connects agents to executable browser workflows rather than static doc fetch alone.
How it compares
Browser task and session MCP, not a lightweight smartfetch-only research server.
Common Questions / FAQ
Who is Web Task API for?
Indie developers using MCP agents who need executable browser tasks and session state, not just URL downloads.
When should I use Web Task API?
During build when integrating agent-driven UI automation, E2E-style checks, or recipe-based flows against live sites.
How do I add Web Task API to my agent?
Install web-task-api from npm, set optional WEB_TASK_API_DATA_DIR and related env vars, add the stdio MCP server to your client, and author or bundle recipes as documented in the repo.