Visual Intent Layer MCP server
io.github.willa-code/visual-intent-layer·v0.4.0Point at visible targets in an agent-built interface, and deliver grounded Annotations.
0 stars99 downloads/wk
Reviews
Write oneNobody has reviewed Visual Intent Layer yet.
If you have run it, two minutes of your experience saves the next person an afternoon.
Visual Intent Layer tools (4)
write = sends, deletes, buys or postsRead from the package source without running it. The installed server may list more.
acknowledge_intentAcknowledge receipt of a batch or of one Annotation as the implementing agent. Acknowledgement is visible to the Builder-Reviewer but never completes the intent and never counts as verification.
check_inCheck in between your own steps to read new direction. Returns everything that arrived since your last check-in: newly delivered Annotations with their delivery intent, a Replacement that replaced something, a pending stop request, and the current state of everything you were given before. This is the Check-In convention: the Builder-Reviewer can amend or interrupt work while you are busy, and it
get_intent_statusRead one delivered batch: each Annotation with its own identity, target resolution (matched, recovered, ambiguous with candidates, or deleted), whether it was written before the revision now on screen, and the agent position. Resolution never auto-selects a guess and acknowledgement never counts as verification.
open_visual_reviewOpen the Visual Direction Loop for an agent-produced web interface and wait for the Builder-Reviewer. Use this when the human\u2019s intent is lossy in words alone: they name a visible thing by location ("that button", "the header", "below the form"), they compare visible things ("align these", "same size as that", "more space here"), a prose-only correction already missed its target once, or they
Public scan report
scanner v0.1.10 · 2026-10-02 · same rubric, same numbers if you re-run it
- Code scan93 source files scanned25/25
- –Live reliabilityno gateway calls yet and no remote to proben/a
- –Tool poisoningtools not inspected (local package is not executed); not countedn/a
- Auth qualitylocal package, no credentials required12/15
- Maintenancelast push 12 days ago15/15
- Maintainer identityregistry namespace matches repository owner6/10
How Visual Intent Layer compares
Against every graded AI and agents MCP server, from our own scans and uptime checks. Updated after each scan run.
- Safety score
- 89/100
- higher than 83% of 4,460 AI and agents servers
- Tools
- 4
- the typical AI and agents server has 6
- Tools that change things
- 0%
- typical: 0%. Fewer write tools means less an agent can break
- Downloads a week
- 99
- more than 64% of 1,538 AI and agents servers with a package
What its tools let an agent do: 3 read · 1 create
Release history: 2 versions in the registry since 2026-09-18, latest 2026-09-20, typically one every 2 days.
Install Visual Intent Layer in Claude Code, Cursor or VS Code
Runs npx -y visual-intent-layer on your machine. Read the scan report first; the gateway never runs local packages.
claude mcp add visual-intent-layer -- npx -y visual-intent-layer
What the publisher says
From the Visual Intent Layer repository's README, as published. We do not edit it. Read it on GitHub
Visual Intent Layer
Point at a browser-rendered interface, describe what should change, and let an agent work from that precise visual context. Visual Intent Layer is a local MCP server and Review Surface for saved HTML documents and local web apps.
About
hello world! this project is a testbed for me to experience new/random/trending stuff! fyi on what i was testing:
- pstack's verification skills
- matt's skills
- ds v4.1 flash via openrouter in pi
- ...
Key features
- Direct pointing. Point, drag, or box visible targets instead of describing their location in prose.
- Annotations with context. Group targets into one Annotation and attach reference images when useful.
- Delivery state you can read. Keep unsent and delivered Annotations together with their revision and resolution state.
- Human judgement. Review the result in a browser and decide each Annotation by hand.
- Local by default. Keep review data on the local machine.
Mobile builds, native windows, design files, and PDFs are outside the supported surface.
Requirements
- Node.js 20 or newer
- A browser on the machine running the server
- An MCP-capable agent, if you want an agent to use the Review Surface
No hosted account or Rust toolchain is required.
Quickstart
Register the local stdio server with your agent, then restart the agent:
{
"mcpServers": {
"visual-intent-layer": {
"command": "npx",
"args": ["-y", "--package", "visual-intent-layer@latest", "visual-intent-mcp"]
}
}
}The server must run on the machine where you look at the screen. It launches a browser there and binds a loopback port. A server in a container, on a remote host, or inside a cloud agent cannot display that machine's Review Surface.
Registry-aware clients can also find the server as io.github.willa-code/visual-intent-layer in the official MCP Registry, which lists where to get it rather than hosting it.
Claude Code
claude mcp add visual-intent-layer -- npx -y --package visual-intent-layer@latest visual-intent-mcpCodex
codex mcp add visual-intent-layer -- npx -y --package visual-intent-layer@latest visual-intent-mcpopencode
Run opencode mcp add visual-intent-layer, then provide npx and the package arguments when prompted:
-y --package visual-intent-layer@latest visual-intent-mcppi
pi does not ship an MCP client. With pi-mcp-adapter installed, add this to the project's .mcp.json (merge it into an existing mcpServers object):
{
"mcpServers": {
"visual-intent-layer": {
"command": "npx",
"args": ["-y", "--package", "visual-intent-layer@latest", "visual-intent-mcp"]
}
}
}The same entry can be placed in .pi/mcp.json for a project-only Pi override, or ~/.config/mcp/mcp.json for all projects. The adapter's /mcp setup panel can also create the entry.
Other MCP clients
Use the JSON configuration above, or launch a local stdio server with command npx and arguments -y --package visual-intent-layer@latest visual-intent-mcp.
Start a review without an agent
The package also provides a local CLI:
npx -y --package visual-intent-layer@latest visual-intent open --html ./checkout.html
npx -y --package visual-intent-layer@latest visual-intent open --app http://localhost:5173Each command opens the default browser and prints the review URL. Add --no-open for headless or scripted use. To run only the local service:
npx -y --package visual-intent-layer@latest visual-intent serve --port 3742The visual-intent command is also available after npm install -g visual-intent-layer.
Shortened. The full README is on GitHub.
Nothing above is checked by us. What we check is on the safety report.
Visual Intent Layer: common questions
- Is Visual Intent Layer MCP server safe?
- Yes, by our scan: it is graded A (89/100). Read the Visual Intent Layer safety report
- How do I install Visual Intent Layer?
- It runs on your machine. Copy the Claude Code, Cursor, VS Code or Claude Desktop config from the install section.
- Does Visual Intent Layer need an API key?
- Not as far as the registry entry and our scan can tell: no credentials are declared or required.
- Is Visual Intent Layer maintained?
- The last commit was 45 days ago (2026-09-20). The latest release is v0.4.0.
- What can I use instead of Visual Intent Layer?
- Servers from other publishers that do the same job: Agent-Native Plan MCP server, agent-device MCP server and Agent Recall MCP server.
- How does Visual Intent Layer compare with other AI and agents MCP servers?
- safety score 89/100 (higher than 83% of 4,460 AI and agents servers); tools 4 (the typical AI and agents server has 6); tools that change things 0% (typical: 0%. Fewer write tools means less an agent can break). Full comparison.
Alternatives to Visual Intent Layer
Same job from other publishers: the closest match first, then the best rated.
- Agent-Native PlanStructured visual plans and PR recaps with diagrams, prototypes, annotations, and sharingnot reviewedEstablishedA
- agent-deviceOpen source device layer for AI agents: control mobile, web, TV, and desktop apps over MCPnot reviewedWidely usedA
- Agent RecallCorrection-first agent memory. Precision KPI tracks if agents heed warnings. 5 layers, local-only.not reviewedEstablishedA
- Agent Security ScannerSecurity layer for AI agents: blocks prompt injection, detects fake packages, scans vulnerabilities.not reviewedEstablishedC