Glasswarp MCP server
com.glasswarp/mcp-serverglasswarp.comstreamable-httpv0.1.0Scanned
Eyes and hands on real Windows PCs — observe, click, type via Glasswarp API.
16 tools: 10 can delete or cancel things and 6 only read.
- Reads private data · 0
- Reads text others write · 14
- Can send data out · 2
- No findings
Tools listed without sign-in. Remote posture
With Anlyon
10 tools here can delete or cancel things. Govern them.
- Anlyon lists the tools and adopts none.
- Each tool you adopt gets approvals and receipts.
- Your agent never holds the key.
A tool your agent calls directly is not routed through Anlyon and is not governed by it. Anlyon Radar only lists and labels, and these labels are the same either way.
What the Glasswarp MCP server can do
The Glasswarp MCP server is published by glasswarp.com, a domain the MCP Registry verified. It has 16 tools. It can delete or cancel things (10 tools: session.start, session.end, input.click_target, input.click_xy and 6 more). 6 tools only read.
14 tools bring in text other people can write (rigs.list, session.start, session.end and 11 more); 2 tools can send data out of your system (input.send_keys and input.send_actions).
None of the fixed checks matched anything in its tool definitions. That is not a safety guarantee.
Its remote endpoint lists its tools without sign-in.
Last scanned Nov 10, 2026, version 0.1.0.
Tools
Grouped by what each tool does, most consequential first. Open one for its reasons and full description; descriptions are the publisher's own words.
16 tools
Deletes or cancels 10Claude Code: deny
app.launchLaunch an executable on the remote Windows rig (name on PATH or absolute path), optional args.Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it destructive (destructiveHint: true).
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Launch an executable on the remote Windows rig (name on PATH or absolute path), optional args. Side effects: starts a process; Glasswarp tracks it and closes it on session.end. Use for notepad.exe, mspaint.exe, chrome with URL args, etc. Wait/re-observe after launch before clicking — do not assume the window is focused immediately.
Annotations the server declares
- destructiveHint: true
- openWorldHint: true
input.click_targetLeft/right/middle-click a UIA target by id from the latest screen.observe (uses native center coords).Reads text others writeNo findings
Why it is in this group
- The name "input.click_target" says "click": a word for a change.
- The server marks it destructive (destructiveHint: true).
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Left/right/middle-click a UIA target by id from the latest screen.observe (uses native center coords). Prefer over input.click_xy. Side effect: real mouse click on the remote Windows desktop. Do not reuse target_id after the screen may have changed — re-observe first. For click→type→keys sequences, use input.send_actions (one turn) instead of chaining this tool.
Annotations the server declares
- destructiveHint: true
- openWorldHint: true
input.click_xyClick at native screen coordinates (0…native_width-1, 0…native_height-1 from screen.observe).Reads text others writeNo findings
Why it is in this group
- The name "input.click_xy" says "click": a word for a change.
- The server marks it destructive (destructiveHint: true).
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Click at native screen coordinates (0…native_width-1, 0…native_height-1 from screen.observe). Last resort when no suitable UIA target exists — prefer input.click_target. Never use JPEG/downscaled pixel coords. Side effect: real mouse click on the remote desktop.
Annotations the server declares
- destructiveHint: true
- openWorldHint: true
input.dragPress-move-release mouse drag in native capture coordinates.Reads text others writeNo findings
Why it is in this group
- The name "input.drag" says "drag": a word for a change.
- The server marks it destructive (destructiveHint: true).
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Press-move-release mouse drag in native capture coordinates. Use for drawing, sliders, selection boxes, and drag-and-drop. Side effect: mouse_down → moves → mouse_up on the remote desktop. Prefer input.send_actions if the drag is one step in a longer predictable sequence.
Annotations the server declares
- destructiveHint: true
- openWorldHint: true
input.scrollMove the cursor to native (x,y) then apply a vertical mouse-wheel delta.Reads text others writeNo findings
Why it is in this group
- The description's first verb is "move": a word for a change.
- The server marks it destructive (destructiveHint: true).
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Move the cursor to native (x,y) then apply a vertical mouse-wheel delta. Side effect: scroll on whatever is under that point. Negative delta scrolls toward the bottom of the page. Prefer input.send_actions when scroll is part of a multi-step sequence. Re-observe after scrolling lists/pages before clicking targets.
Annotations the server declares
- destructiveHint: true
- openWorldHint: true
input.send_actionsPREFERRED multi-step tool: run 1–10 predictable UI actions in one call (input.click_target, input.click_xy, input.type_text, input.send_keys, input.drag, input.scroll).Reads text others writeCan send data outNo findings
Why it is in this group
- The name "input.send_actions" says "send": a word for sending, and what is sent cannot be taken back.
- The server marks it destructive (destructiveHint: true).
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
- outboundThe name "input.send_actions" says "send": it sends or publishes something to other people.
Description, as the server wrote it PREFERRED multi-step tool: run 1–10 predictable UI actions in one call (input.click_target, input.click_xy, input.type_text, input.send_keys, input.drag, input.scroll). Side effects: all actions execute on the remote desktop; fails fast before sending if any action is invalid. observe_after defaults true (verification observe: text+targets; set observe_image=true for JPEG). Do not batch across unpredictable waits (page loads, installers, modals) — single-step those. Prefer this over chaining solo click/type/keys tools.
Annotations the server declares
- destructiveHint: true
- openWorldHint: true
input.send_keysSend a key or chord to the focused window (e.g.Reads text others writeCan send data outNo findings
Why it is in this group
- The name "input.send_keys" says "send": a word for sending, and what is sent cannot be taken back.
- The server marks it destructive (destructiveHint: true).
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
- outboundThe name "input.send_keys" says "send": it sends or publishes something to other people.
Description, as the server wrote it Send a key or chord to the focused window (e.g. enter, tab, ctrl+s, alt+f4, win). Side effect: real key events on the remote desktop. Prefer bundling into input.send_actions when the shortcut follows a click/type in the same planned sequence. Use input.type_text for literal strings, not this tool.
Annotations the server declares
- destructiveHint: true
- openWorldHint: true
input.type_textType a Unicode string into the currently focused control via native input.Reads text others writeNo findings
Why it is in this group
- The name "input.type_text" says "type": a word for a change.
- The server marks it destructive (destructiveHint: true).
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Type a Unicode string into the currently focused control via native input. Does not click first — focus the field (input.click_target / input.send_actions) before calling. Side effect: keystrokes on the remote desktop. For form fills (click → type → tab/enter), prefer input.send_actions in one call.
Annotations the server declares
- destructiveHint: true
- openWorldHint: true
session.endEnd an active session.Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it destructive (destructiveHint: true).
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it End an active session. Side effects: stops billing, runs host safety_restore, closes apps launched via app.launch. Always call when finished or abandoning — do not leave sessions open. Safe to call once; further observe/input on that session_id will fail.
Annotations the server declares
- destructiveHint: true
- openWorldHint: true
session.startStart a metered desktop session on a USABLE rig from rigs.list.Reads text others writeNo findings
Why it is in this group
- The name "session.start" says "start": a word for a change.
- The server marks it destructive (destructiveHint: true).
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Start a metered desktop session on a USABLE rig from rigs.list. Side effects: begins wall-clock billing, shows an on-screen “API session active” indicator, enables observe/input until session.end. Idle sessions auto-end after ~15 minutes. Always call session.end when done or abandoning. Do not call if no USABLE rig exists. Returns session_id and Live View URL (owner console login required).
Annotations the server declares
- destructiveHint: true
- openWorldHint: true
Only reads 6Show 6 read-only tools
demos.getReturn one showcase run contract (install, command, needs, framing).No findings
Why it is in this group
- The name "demos.get" says "get": a word for reading.
Description, as the server wrote it Return one showcase run contract (install, command, needs, framing). Does NOT execute the demo or control a PC. Call after demos.list when you know the demo_id. If the client can run shell, offer the command; if chat-only, show the card. Do not replace this with a slow MCP click loop for solver demos.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
demos.listList showcase run contracts (id, title, install, command).No findings
Why it is in this group
- The name "demos.list" says "list": a word for reading.
Description, as the server wrote it List showcase run contracts (id, title, install, command). Read-only catalog — does NOT start a session, touch a rig, or run solvers. Use when the user asks for Minesweeper/Mona Lisa/Paint demos or you need the glasswarp-demo command. Prefer demos.get for one full card. For ad-hoc UI work use rigs.list → session.start → screen.observe instead.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
rigs.listList Windows machines (rigs) paired to this API key: id, name, online, api_access_enabled, and USABLE flag.Reads text others writeNo findings
Why it is in this group
- The name "rigs.list" says "list": a word for reading.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it List Windows machines (rigs) paired to this API key: id, name, online, api_access_enabled, and USABLE flag. Read-only — does not start a session. Call first before start_session. A rig is USABLE only when online AND the owner enabled API access. If none are USABLE, tell the user to install the host agent, pair in Console → Rigs, and enable API access — never ask for OS passwords.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: true
screen.observeRead the current screen: UIA targets (numbered ids + native coords) and a text summary.Reads text others writeNo findings
Why it is in this group
- The description's first verb is "read": a word for reading.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Read the current screen: UIA targets (numbered ids + native coords) and a text summary. Does not move mouse/keyboard. Default image=false (no JPEG) for speed; set image=true only when you must judge pixels visually (then max_width≈960, quality≈60). If changed=false, JPEG is omitted even when requested — do not re-analyze; wait or act differently. If dirty is null, assume changed. Prefer input.send_actions for multi-step UI; observe after meaningful steps, not after every click. Target ids are valid only until the next UI change.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: true
session.live_viewReturn the console Live View URL (≈60fps) for the rig owner to watch and intervene.Reads text others writeNo findings
Why it is in this group
- The name "session.live_view" says "view": a word for reading.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
- untrusted-inputThe description's first sentence says "return" and "url": it reads web pages, which anyone can write.
Description, as the server wrote it Return the console Live View URL (≈60fps) for the rig owner to watch and intervene. Read-only for the agent — does not grant the API key console access. Offer on long or sensitive tasks. Owner must be signed into Glasswarp; API keys alone cannot open the player.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: true
session.statusFetch session metadata: status, host, mode, created_at, action_count, billed_minutes.Reads text others writeNo findings
Why it is in this group
- The name "session.status" says "status": a word for reading.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Fetch session metadata: status, host, mode, created_at, action_count, billed_minutes. Read-only — no input side effects. Use to tell the user about metered time or confirm the session is still active before more actions.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: true
This is not a safety rating, and no finding is not a safety guarantee. The groups and findings come from fixed checks of the tool definitions the server returned on the date shown, not from its code or what it does when called. How the checks work.
Exfil surface
Which tools read private data, which read text someone else can write, and which can send data out. One of each is a path a prompt injection can use.
No private-data, untrusted-input, outbound path in this server alone: no tool is tagged private-data.
- Reads private data0
- Reads files, mail, messages, databases, secrets, calendars or repositories.
- Reads text others write14
- Brings in text someone else can write: web pages, inbound email, messages, issues and comments.
- Can send data out2
- Can carry data to people or systems: it sends, posts or publishes, or takes a recipient, channel or URL.
Other servers in the same agent add their own tools to these paths. Audit your whole setup to see them together.
Remote posture
What anyone on the internet sees at the remote endpoint: an anonymous initialize and tools/list, and the public OAuth metadata.
Looked at Nov 20, 2026, 18:39 UTC, unchanged since Oct 22, 2026. No credentials were sent and no tool was called. What is looked at.
- Reachable
- yes
- HTTP status of an anonymous initialize
- 200
- tools/list served without credentials
- yes
- Tools that change things, listed without credentials
- 10
- Protocol version
- 2025-06-18
- WWW-Authenticate on 401
- not observed
- resource_metadata in WWW-Authenticate
- not observed
- RFC 9728 protected resource metadata
- not observed
- Authorization servers
- not observed
- Legacy HTTP+SSE transport
- no
- Dynamic client registration
- not observed
- Client ID metadata documents
- not observed
- HighThe whole server
The endpoint served tools/list without credentials, and the list has 10 tools that change things (10 destructive). Whether a call also works without credentials was not tested: tools are never called.
What this check looks forMatched text
https://mcp.glasswarp.com/mcp: tools/list returned 16 tools to an anonymous client, including session.start (destructive), session.end (destructive), input.click_target (destructive), input.click_xy (destructive), input.type_text (destructive) and 5 more
Findings
Matches of fixed checks, with the text they matched.
No check matched anything in these tool definitions. That is not a safety guarantee.
History
Changes to its tool definitions and to its remote endpoint, newest first.
No change seen yet. Its tools are rescanned every day, and a remote endpoint is looked at once a week.
Install
Pinned to version 0.1.0. Values in angle brackets are yours to fill in.
claude mcp add --transport http glasswarp https://mcp.glasswarp.com/mcp --header 'Authorization: <your-authorization>'claude mcp add --transport http glasswarp https://mcp.glasswarp.com/mcp --header 'Authorization: <your-authorization>'Run in a terminal.
Claude Code permissions
Allow the reads, ask before a change or a way out, deny the deletes. Paste into .claude/settings.json.
{
"permissions": {
"allow": [
"mcp__glasswarp__demos_list",
"mcp__glasswarp__demos_get",
"mcp__glasswarp__rigs_list",
"mcp__glasswarp__screen_observe",
"mcp__glasswarp__session_live_view",
"mcp__glasswarp__session_status"
],
"ask": [],
"deny": [
"mcp__glasswarp__session_start",
"mcp__glasswarp__session_end",
"mcp__glasswarp__input_click_target",
"mcp__glasswarp__input_click_xy",
"mcp__glasswarp__input_type_text",
"mcp__glasswarp__input_send_keys",
"mcp__glasswarp__input_drag",
"mcp__glasswarp__input_scroll",
"mcp__glasswarp__input_send_actions",
"mcp__glasswarp__app_launch"
]
}
}Tools that only read are allowed, tools that change things ask first, tools that delete or cancel are denied. A read that can send data out (it takes a URL, say) asks first too. It assumes the server is installed as glasswarp. Review it before you use it.
Capability tags behind 14 rules
- mcp__glasswarp__rigs_list untrusted-input
- mcp__glasswarp__session_start untrusted-input
- mcp__glasswarp__session_end untrusted-input
- mcp__glasswarp__screen_observe untrusted-input
- mcp__glasswarp__input_click_target untrusted-input
- mcp__glasswarp__input_click_xy untrusted-input
- mcp__glasswarp__input_type_text untrusted-input
- mcp__glasswarp__input_send_keys untrusted-input, outbound
- mcp__glasswarp__input_drag untrusted-input
- mcp__glasswarp__input_scroll untrusted-input
- mcp__glasswarp__input_send_actions untrusted-input, outbound
- mcp__glasswarp__app_launch untrusted-input
- mcp__glasswarp__session_live_view untrusted-input
- mcp__glasswarp__session_status untrusted-input
Pin
Keep the decision: fail CI when the tools change, follow every change, or show a badge.
Pin in CI
Fail a build when the tools you reviewed are not the tools the server serves today.
curl -s "https://radar.anlyon.com/api/v1/servers/com.glasswarp/mcp-server/verify?hash=1622199258d78a5f1c5ff2171ae9d2338e7b73bb4cf37514d5a32d46d46a612f" | jq -e '.changed == false'Pinned hash 1622199258d7, from Nov 10, 2026, 16:18 UTC.
Watch
Get told when its tools change: a new tool, a new description, a new class.
https://radar.anlyon.com/servers/com.glasswarp/mcp-server/changes.xmlAn Atom feed. Works in any feed reader or Slack's RSS app.
Badge
[](https://radar.anlyon.com/servers/com.glasswarp/mcp-server)Source: MCP Registry · repository. Synced Oct 21, 2026, 17:55 UTC.
Glasswarp MCP server: questions and answers
Answered from the scan above, not from the publisher's description.
Is the Glasswarp MCP server safe?
- No scan can promise that. What its definitions show: 16 tools, 10 of which can change, send or delete something, and 0 findings from fixed checks. Allow the reads, ask before each change, and deny the deletes; the permissions block on this page does that.
Can the Glasswarp MCP server delete data?
- Yes. 10 tools can delete, cancel or revoke something: session.start, session.end, input.click_target, input.click_xy, input.type_text, input.send_keys and 4 more.
Can the Glasswarp MCP server do things that cannot be undone?
- No tool is classed as irreversible: none of them sends, pays, publishes or merges, going by their definitions.
What tools does the Glasswarp MCP server have?
- 16 tools: 10 destructive, 6 read. session.start, session.end, input.click_target, input.click_xy, input.type_text, input.send_keys, input.drag, input.scroll and 8 more.
Does the Glasswarp MCP server need sign-in or OAuth?
- Its remote endpoint listed its tools to an anonymous client, without sign-in.
Does the Glasswarp MCP server send data to outside services?
- 2 tools can carry data out (it sends, posts or publishes, or takes a recipient, channel or URL): input.send_keys and input.send_actions.
Has the Glasswarp MCP server changed its tools?
- Not since Anlyon Radar started tracking it. Last scanned Nov 10, 2026.