Atlas / MCP servers / ageborn-dev / Architect

ArchitectBLOCK

Name: Architect
Author: ageborn-dev

mcp/ageborn-dev/architect-1

MCP server by ageborn-dev · ai · ai agents · aitools · Audited 2026-10-09

A powerful, self-extending MCP server for dynamic AI tool orchestration. Features sandboxed JS execution, capability-based security, automated rate limiting, marketplace integration, and a built-in monitoring dashboard. Built for the Model Context Protocol (MCP).

Verdict

BLOCK

Grade

Trust score

69 /100

Exposed tools

84 62r · 17w · 5d

Transport

stdio

License

MIT

Stars

Overview

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

The AI agent workshop that builds its own tools.

🚀 What is Architect?

AI agents are incredibly smart, but they hit a wall when they need a tool that doesn't exist yet. Architect MCP removes that wall.

Instead of giving your AI agent a fixed, rigid toolbox, you give it an entire workshop. If your agent needs to call an unconventional API, parse an obscure file format, or hook into a database, it simply writes a custom tool for it on the fly.

Once approved by you, that tool runs securely in an isolated sandbox. The next time your agent hits the same problem? The customized tool is already there, ready to go.

✨ Key Features

Automated Tool Creation: Agents construct, test, and permanently save their own JavaScript tools on the fly.
Ironclad Security & Sandboxing: Every custom tool is executed in a highly restricted sandbox. Network or file system access requires explicit user approval through granular capability scopes.
Built-in Web Dashboard: A beautiful, real-time dashboard UI (running on port 3001 out of the box). Manage your active tools, watch execution logs unfold in real time, monitor failures, and securely manage secrets.
Cron Scheduling: Not just for manual use! Agents can set up custom tools to run on cron schedules or build continuous background pipelines.
Global Marketplace: Why build from scratch if someone else already did? Agents can search (marketplace_browse), install (marketplace_install), and even share your creations (marketplace_publish) using a GitHub token.
Persistent Data Layer: Built-in, blazing-fast SQLite storage to manage tools, run logs, and execution states reliably.

🛠️ Getting Started

1. Simple Local Setup

You can get running in seconds if you have Node.js installed.

npm install
npm run build
npm start

Or once pub

Read from source at commit 191d84914476OBSERVED · 2026-10-09

Connect

Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control. Replace the environment placeholders with a token scoped to the least it needs.

claude-code

claude mcp add architect-mcp-server --env DASHBOARD_SECRET=${DASHBOARD_SECRET} -- npx -y [email protected]

claude-desktop

{
  "mcpServers": {
    "architect-mcp-server": {
      "command": "npx",
      "args": [
        "-y",
        "[email protected]"
      ],
      "env": {
        "DASHBOARD_SECRET": "${DASHBOARD_SECRET}"
      }
    }
  }
}

Exposed tools (84)

62 read · 17 write · 5 destructive. Blast radius: 5 tools can delete or overwrite — an agent that can be talked into calling a tool can be talked into calling this one.

Tool	Risk	Description
`browser_auth_login`	read	Login using saved credentials from the auth vault.
`browser_back`	read	Go back in browser history.
`browser_check`	read	Check a checkbox.
`browser_click`	read	Click an element. Use @ref from browser_snapshot (e.g. @e2) or CSS/text selectors.
`browser_close`	read	Close the browser and end the session.
`browser_connect`	read	Connect to an existing Chrome browser via CDP (Chrome DevTools Protocol) port.
`browser_console`	read	Read browser console messages (console.log, warn, error, info).
`browser_cookies_clear`	destructive	Clear all cookies.
`browser_cookies_get`	read	Get all cookies for the current session.
`browser_cookies_set`	write	Set a cookie.
`browser_dblclick`	read	Double-click an element.
`browser_dialog_accept`	read	Accept (OK) a browser dialog (alert, confirm, prompt).
`browser_dialog_dismiss`	read	Dismiss (Cancel) a browser dialog.
`browser_diff_screenshot`	read	Visual pixel diff between current screenshot and a baseline image.
`browser_diff_snapshot`	read	Compare the current accessibility tree snapshot against a previous one to detect page changes.
`browser_diff_url`	read	Compare two URLs side-by-side via snapshot diff (and optionally screenshot diff).
`browser_drag`	read	Drag an element from source to target.
`browser_eval`	write	Execute JavaScript in the context of the current page and return the result.
`browser_fill`	destructive	Clear and fill an input field. Preferred over browser_type for form inputs.
`browser_find`	read	Find elements using semantic locators (by ARIA role, text, label, placeholder, alt, testid, nth). More robust than CSS selectors.
`browser_focus`	read	Focus an element without clicking it.
`browser_forward`	read	Go forward in browser history.
`browser_frame`	read	Switch context to an iframe, or back to the main frame.
`browser_get_attr`	read	Get the value of an HTML attribute on an element.
`browser_get_box`	read	Get the bounding box (x, y, width, height) of an element.
`browser_get_count`	read	Count how many elements match a selector.
`browser_get_html`	read	Get the innerHTML of an element.
`browser_get_styles`	read	Get the computed CSS styles of an element.
`browser_get_text`	read	Get the visible text content of an element.
`browser_get_title`	read	Get the current page title.
`browser_get_url`	read	Get the current page URL.
`browser_get_value`	read	Get the current value of an input element.
`browser_highlight`	read	Visually highlight an element on the page (useful for debugging in headed mode).
`browser_hover`	read	Hover over an element (useful for revealing tooltips or dropdown menus).
`browser_is_checked`	read	Check whether a checkbox or radio button is checked.
`browser_is_enabled`	read	Check whether an element is enabled (not disabled).
`browser_is_visible`	read	Check whether an element is visible on the page.
`browser_keyboard_insert_text`	write	Insert text without triggering key events (no element selector). Faster than keyboard_type.
`browser_keyboard_type`	read	Type text using real keystrokes (no element selector, types at current focus).
`browser_keydown`	read	Hold a key down.
`browser_keyup`	read	Release a held key.
`browser_mouse_down`	read	Press a mouse button down.
`browser_mouse_move`	write	Move the mouse cursor to specific coordinates.
`browser_mouse_up`	read	Release a pressed mouse button.
`browser_mouse_wheel`	read	Scroll the mouse wheel.
`browser_network_requests`	read	View tracked network requests made by the page.
`browser_network_route`	read	Intercept, block, or mock network requests matching a URL pattern.
`browser_network_unroute`	destructive	Remove a network route/intercept.
`browser_open`	read	Navigate to a URL. Returns the page after load. Use browser_snapshot after to see interactive elements.
`browser_page_errors`	read	Read uncaught JavaScript exceptions on the page.
`browser_pdf`	write	Save the current page as a PDF file.
`browser_press`	read	Press a keyboard key or shortcut (e.g. Enter, Tab, Control+a, Escape).
`browser_reload`	read	Reload the current page.
`browser_screenshot`	read	Take a screenshot. Use --annotate to get numbered labels matching @refs for visual+text workflows.
`browser_scroll`	read	Scroll the page or a specific element in a direction.
`browser_scroll_into_view`	read	Scroll a specific element into the viewport.
`browser_select`	read	Select an option in a <select> dropdown.
`browser_session_list`	read	List all active isolated browser sessions.
`browser_set_credentials`	write	Set HTTP basic auth credentials.
`browser_set_device`	write	Emulate a device (e.g.
`browser_set_geo`	write	Set the browser
`browser_set_headers`	write	Set global HTTP headers for all requests (all domains).
`browser_set_media`	write	Emulate a CSS color scheme preference.
`browser_set_offline`	write	Toggle offline mode on or off.
`browser_set_viewport`	write	Set the browser viewport size.
`browser_snapshot`	read	Get the current page accessibility tree with @ref handles for interacting with elements. Best first step after navigation.
`browser_state_clear`	destructive	Clear saved browser auth state(s).
`browser_state_list`	read	List all saved browser auth state files.
`browser_state_load`	read	Load a previously saved browser auth state from a file.
`browser_state_save`	write	Save current browser auth state (cookies, localStorage) to a file for reuse.
`browser_storage_clear`	destructive	Clear all values from localStorage or sessionStorage.
`browser_storage_get`	read	Get value(s) from localStorage or sessionStorage.
`browser_storage_set`	write	Set a value in localStorage or sessionStorage.
`browser_tab_close`	read	Close a browser tab.
`browser_tab_list`	read	List all open browser tabs.
`browser_tab_new`	read	Open a new browser tab, optionally navigating to a URL.
`browser_tab_switch`	read	Switch to a tab by its index number.
`browser_trace_start`	write	Start recording a Playwright trace for debugging and replay.
`browser_trace_stop`	write	Stop and save the current Playwright trace.
`browser_type`	read	Type text into an element (appends to existing content).
`browser_uncheck`	read	Uncheck a checkbox.
`browser_upload`	write	Upload one or more files to a file input element.
`browser_wait`	read	Wait for an element, text, URL pattern, load state, or a time duration before continuing.
`browser_window_new`	read	Open a new browser window.

Trust audit

BLOCKgrade D · trust 69/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.

Layer	What it checks	Result
L0	Provenance & inventory	PASS
L1	Static analysis of the code	FAIL
L2	Instruction surface (what it tells the agent)	PASS
L3	Class-specific surface	WARN
L4	Behavioural (sandbox)	SKIPPED

What the source does

Filesystem: declared (2 observation(s))
Network: declared (5 observation(s))
Shell: declared (2 observation(s))
Dependencies: not all pinned
Secrets in source: none-found

Findings (6)

HIGHCode injection · code.eval_exec · CWE-78, CWE-94, CWE-95

src/execution/pipelines.ts:129

const condResult = new Function("ctx", `return ${step.condition}`)(context);

Why it matters. evaluates text as code

Fix. remove; use a parser or a dispatch table

HIGHCode injection · code.eval_exec · CWE-78, CWE-94, CWE-95

src/tools/templates.ts:174

const result = await exec(params.command, args);

Why it matters. evaluates text as code

Fix. remove; use a parser or a dispatch table

MEDIUMFilesystem / path · mcp.destructive_tools · CWE-22, CWE-59

browser_cookies_clear, browser_fill, browser_network_unroute, browser_state_clear, browser_storage_clear

Why it matters. 5 tool(s) can delete or overwrite

Fix. prefer a read-only mode or scoped tokens; the page states the blast radius

LOWFilesystem / path · fs.traversal · CWE-22, CWE-59

src/core/sandbox-process.ts:43

const require = createRequire(path.resolve(__dirname, "../../node_modules"));

LOWSupply chain · supply.unpinned · CWE-829, CWE-1357

package.json

@hono/node-server, @modelcontextprotocol/sdk, better-sqlite3, cron-parser, hono, zod, @types/better-sqlite3, @types/node

Why it matters. 10 dependency range(s) float

Fix. pin exact versions or ship a lockfile

INFOPrompt injection · prompt.fetch_and_trust · CWE-94, CWE-1427

CONTRIBUTING.md:35

The server will start and the dashboard will be at `http://localhost:3001`. The `npm run dev` command watches for changes and restarts automatically, so you don't need to rebuild manually while workin

Why it matters. remote text is to be obeyed as instructions

Gates applied: no_behavioural_pass.

Audited 2026-10-09 · audit v0.4.1 · source sha 191d84914476full audit observations/trust-audit/mcp-server/ageborn-dev__architect-1.json · Report an issue / request a re-scan

Audit history

Every audit this server has had. A grade with a past is a grade somebody is still checking.

Date	Source	Verdict	Grade	Score	Change
2026-10-09	`191d84914476`	BLOCK	D	69	first audit

Questions

What is the Architect MCP server?

What tools does Architect expose?

84 in total: 62 read-only, 17 that write, and 5 that can delete or overwrite (browser_cookies_clear, browser_fill, browser_network_unroute, browser_state_clear, browser_storage_clear). Every one is listed on this page with its risk.

Is Architect safe to connect to an agent?

No — not without reading the findings first. The audit graded it D (69/100) and found 2 critical or high issues in the source. Each one is listed on this page with the file and line it is on. Separately from the audit: 5 of its tools can destroy data, so scope the token you give it to what you actually need.

What credentials does Architect need?

It reads DASHBOARD_SECRET from the environment. Give it a token scoped to the least it needs — an agent that can be talked into calling a tool can be talked into calling it with your credentials.

How does Architect run?

It speaks stdio, so it runs as a local process your client starts. It is published on npm as architect-mcp-server at 1.0.0.

How current is this page?

The grade is for one exact copy of the source (191d84914476), read on 2026-10-09. The repository is watched and re-audited when it changes.

Source	ageborn-dev/architect-mcp-server
npm	`architect-mcp-server@1.0.0`
Transports	stdio
Credentials it reads	`DASHBOARD_SECRET`
License	MIT
Activity	pushed 118d ago