About Klura
Klura is an MCP server that exposes local web-data tools and browser automation through the klura runtime. It is for agents that need to search, run, or explicitly author tools while keeping target traffic and outputs on the user's machine.
Key features
- 01Search and install signed local tools
- 02Call and run tool capabilities
- 03Automate browser sessions on request
- 04Inspect network logs and page content
- 05Read reference material through a resource
Use cases
- 01Search and run local tools for web data
- 02Automate browser actions when explicitly requested
- 03Inspect page accessibility and network activity
MCP support
The server mirrors klura runtime functions as MCP tools and serves the `klura://reference` resource. It communicates over stdio using JSON-RPC.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
Tools
Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.
Details
From the project README@klura/mcp
MCP (Model Context Protocol) server for @klura/runtime. It exposes maintained local web-data tools first, with browser discovery and tool authoring available only when the user explicitly asks for them.
@klura/runtime searches, installs, calls, and runs signed local tools without a model key. Its separate factory can map and maintain tools through observed browser work. @klura/mcp is the thin wrapper that speaks MCP on top of both surfaces.
Install
npm install -g @klura/mcp
The @klura/runtime package is declared as a dependency and installed alongside. The runtime auto-starts a local daemon on first use and stores everything it learns under ~/.klura/.
Wire it up
Add the server to your MCP client's config. The exact file path depends on the client — the server definition is the same.
{
"mcpServers": {
"klura": {
"command": "npx",
"args": ["-y", "@klura/mcp"]
}
}
}
Restart the client. The agent picks up the klura toolset automatically.
What it exposes
Two surfaces land in the agent's context:
- Maintained local tools — search, inspect, install, call, run, and inspect a signed package through
search_packages,install_package,call_package_capability,start_scrape_run, and the related run/session tools. Target traffic and outputs stay on the user's machine. - Explicit authoring — browser automation (
start_session,perform_action,get_screenshot,get_a11y_tree), discovery + persistence (save_strategy,execute,list_platform_skills,get_strategy), network-log inspection (get_network_log,find_in_page), and reverse-engineering tools. The runtime owns the canonical list; this server mirrors it one-for-one. - Resource
klura://reference— the detailed reference doc, served section-by-section via URL fragments (klura://reference#reverse-engineer-playbook,klura://reference#recorded-path-schema, etc.) so each response fits inside the MCP output budget. Fetchklura://referencewith no fragment for a table of contents.
The always-loaded orientation is SKILL.md, passed as the server's instructions capability. Agents read SKILL.md on every conversation and pull detail on demand via the klura://reference fragments.
How it works
Thin wrapper — each MCP tools/call dispatches to the corresponding klura runtime function. The runtime handles daemon lifecycle, browser sessions, strategy persistence under ~/.klura/skills/, and execution.
MCP client (Claude Desktop, Cursor, …)
│ stdio transport, JSON-RPC
▼
@klura/mcp (this package)
│ Node require('@klura/runtime')
▼
klura runtime → local daemon → Playwright
License
BUSL-1.1. See LICENSE. Same terms as the underlying @klura/runtime; see the runtime README for the commercial-use terms.
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.