PanoOn Street View

PanoOn MCP + agent skills

Give any MCP-capable code agent the ability to scan and download Street View data through PanoOn: mcp/panoon-mcp.mjs is a zero-dependency MCP stdio server that wraps the PanoOn local service and the distributed controller.

Quick start (npm)

npx -y panoon-mcp

The npm package panoon-mcp exposes the panoon-mcp bin. You can also download panoon-mcp.zip from https://app.rjfan.org/panoon/mcp/, or use the copy bundled with PanoOn 0.2.21+ (<install>\mcp\panoon-mcp.mjs).

Prerequisite: the PanoOn app/service is running on the same machine (http://127.0.0.1:8787; the desktop app starts it automatically since 0.2.20).

The server

node mcp/panoon-mcp.mjs# or with the runtime bundled with PanoOn (no Node install required):"C:\Users\<you>\AppData\Local\PanoOn\panoon-service-runtime.exe" mcp\panoon-mcp.mjs

Environment:

Tools: panoon_status, panoon_area_scan, panoon_area_status, panoon_area_result, panoon_area_cancel, panoon_download, panoon_job_status, panoon_jobs, panoon_workers, panoon_cancel_job.

Install per agent

OpenCode (V2)

opencode mcp add panoon --global -- node <repo>\mcp\panoon-mcp.mjsopencode mcp list

or by hand in ~/.config/opencode/opencode.jsonc:

{  "mcp": {    "servers": {      "panoon": {        "type": "local",        "command": ["node", "<repo>/mcp/panoon-mcp.mjs"],        "environment": { "PANOON_SERVICE_URL": "http://127.0.0.1:8787" }      }    }  }}

Skill: copy mcp/skills/panoon/ into your OpenCode skills location (or reference the folder from the skills configuration field).

Claude Code

claude mcp add --scope user panoon -- node <repo>/mcp/panoon-mcp.mjs

or .mcp.json:

{ "mcpServers": { "panoon": { "command": "node", "args": ["<repo>/mcp/panoon-mcp.mjs"] } } }

Skill: copy mcp/skills/panoon/ to ~/.claude/skills/panoon/.

Codex CLI

Add to ~/.codex/config.toml:

[mcp_servers.panoon]command = "node"args = ["<repo>/mcp/panoon-mcp.mjs"]env = { PANOON_SERVICE_URL = "http://127.0.0.1:8787" }

For Codex, paste the workflow section of mcp/skills/panoon/SKILL.md into your AGENTS.md so the agent knows the intended flow.

Any other MCP client (stdio)

Use the generic stdio contract:

{  "command": "node",  "args": ["<repo>/mcp/panoon-mcp.mjs"],  "env": { "PANOON_SERVICE_URL": "http://127.0.0.1:8787" }}

DeepSeek Harness (dsh)

deepseek mcp add panoon --command node --arg <repo>\mcp\panoon-mcp.mjsdeepseek mcp list

This writes ~/.deepseek/mcp.json (servers.panoon; tools surface namespaced by the harness). Install the skill by copying mcp/skills/panoon/ into ~/.deepseek/skills/panoon/ — dsh uses the same SKILL.md contract as Claude Code — then restart the harness so both are picked up.

On this machine the working example is:

deepseek mcp add panoon --command node --arg D:\04_Dev\PanoOn\mcp\panoon-mcp.mjscopy D:\04_Dev\PanoOn\mcp\skills\panoon %USERPROFILE%\.deepseek\skills\panoon

Typical agent flow

1. panoon_status → service healthy, strategies available.
2. panoon_area_scan { polygon: [{lat,lng}, ...] } → jobId; panoon_area_result → PanoIDs.
3. Confirm output directory with the user.
4. panoon_download { panoids, outputDir } → jobId; panoon_job_status for progress.

Safety: downloads are large (~0.5 MB/image) and consume proxy quota. Agents should confirm scope and output paths before submitting jobs; scans are cheap and can be cancelled at any time.