Skip to main content
Glama
@sentiac/mcp

@sentiac/mcp

Official
by Sentiac

Sentiac agent tools

CI npm License: MIT

One MCP server and a set of skills that let a coding agent (Claude Code, Codex, OpenCode, Pi, Cursor, …) use Sentiac tools.

What it is

Where

@sentiac/mcp

The MCP server (stdio). Toolsets are loaded by name; today: Visor, the browser viewer for ROS 2 robots, driven by your agent.

packages/mcp

Skills

Agent Skills (SKILL.md) that teach an agent a workflow.

skills/

Claude Code plugin

The same repo is a plugin: skills + MCP server in one install.

.claude-plugin/, .mcp.json

Skills today: sentiac-visor (draw any robot data with recipes), sentiac-visor-nav2 (a Nav2 view), sentiac-visor-moveit2 (a MoveIt 2 arm view).

Install

The quickest way, for every agent you have installed (it shows what it will change and asks first):

npx -y @sentiac/mcp init

init adds the MCP server in each agent's own way (claude mcp add, codex mcp add, opencode.json, ~/.cursor/mcp.json) and copies the skills to where that agent reads them. Run npx -y @sentiac/mcp uninstall to undo it, npx -y @sentiac/mcp doctor to check things.

Or by hand, per agent:

Agent

MCP server

Skills

Claude Code

claude mcp add --scope user sentiac -- npx -y @sentiac/mcp@^0.1.0

/plugin marketplace add Sentiac/agent-tools then /plugin install sentiac@sentiac-tools (brings both), or npx skills add Sentiac/agent-tools

Codex

codex mcp add sentiac -- npx -y @sentiac/mcp@^0.1.0

npx skills add Sentiac/agent-tools (installs to ~/.agents/skills)

OpenCode

opencode.json: "mcp": { "sentiac": { "type": "local", "command": ["npx", "-y", "@sentiac/mcp@^0.1.0"] } }

npx skills add Sentiac/agent-tools

Cursor

~/.cursor/mcp.json: "mcpServers": { "sentiac": { "command": "npx", "args": ["-y", "@sentiac/mcp@^0.1.0"] } }

Pi

no built-in MCP client (use an MCP extension)

npx skills add Sentiac/agent-tools -a pi

Anything else

any client that can start a stdio server: npx -y @sentiac/mcp

copy a folder from skills/

Only some toolsets: npx -y @sentiac/mcp --tools visor (or SENTIAC_TOOLS=visor). Every tool description costs context in every conversation, so load only what you use.

Related MCP server: AKIMCP

Using Visor with your agent

  1. Open Visor in your browser, then ask your agent to "link Visor". It gives you a short code; enter it in the Link your agent box that appears at the bottom right of Visor (once; Visor remembers it). Both must be on the same computer. Chrome asks once whether Visor may connect to this computer (it only ever talks to your own machine); choose Allow. If the box does not show up, open Settings, Agent, and press Look again.

  2. Start your robot's Visor bridge and give the agent the connection code it prints.

  3. Ask for what you want to see: "show the lidar and the map", "plot the odometry speed", "set up a Nav2 view".

The MCP server runs on your machine and talks to your own browser tab over a local, token-protected link, so the agent sees and changes what you see. It cannot write to the robot. Scripts an agent writes run in a sandbox in your browser, and you are asked once before the first one runs.

Develop and test locally (before anything is published)

git clone https://github.com/Sentiac/agent-tools && cd agent-tools
npm install
npm test            # unit, end-to-end (stdio server + a fake page), installer, and the packed tarball
npm run check       # syntax, skills (strict YAML, names, referenced tools), versions, vendored files

Try it in an agent without publishing:

npm run local:plugin            # builds .local/plugin: the same plugin, running this checkout with node
claude --plugin-dir .local/plugin
node packages/mcp/bin/sentiac-mcp.mjs init --local     # or set up every agent to run this checkout

Any agent that reads Claude-style plugin folders (.claude-plugin/, skills/, .mcp.json) can load .local/plugin the same way.

More in docs/DEVELOPING.md (layout, how the Visor library is brought in, adding a toolset or a skill) and docs/PUBLISHING.md (what to settle and do before the first release).

Contributing

Issues and pull requests are welcome; see CONTRIBUTING.md. Report security problems privately, as described in SECURITY.md. Changes are listed in CHANGELOG.md.

License

MIT © 2026 Sentiac Private Limited

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables AI agents to send HTML documents to a local viewer, where they are stored and displayed in a live web UI for browsing and searching.
      9 npm
      ISC
    • F
      license
      Not graded
      quality
      A
      maintenance
      Exposes an OAuth-gated suite of 38 tools that let web-based AI chat clients securely operate a local machine, covering files, shell, search, Git, SQLite, browser automation, DevTools, background tasks, localhost services, clipboard, notifications, ports, and Postman. A strict command whitelist and configurable folder allowlist keep that broad local access governed rather than unrestricted.
      -
    • A
      license
      Not graded
      quality
      A
      maintenance
      Exposes a governed suite of 38 tools over a single OAuth-gated MCP endpoint, letting web-based AI chat clients read and edit local files, run whitelisted shell commands, search, use Git, inspect SQLite databases, and drive browser, DevTools, clipboard, and background-task workflows. It requires no system-wide install or admin rights and restricts access to folders and commands approved from a local control panel.
      MIT
    • A
      license
      Not graded
      quality
      A
      maintenance
      Exposes an OAuth-gated suite of 38 tools that let web-based AI chat clients operate a local machine, covering files, whitelisted shell commands, search, Git, SQLite, browser and DevTools automation, background tasks, localhost services, clipboard, and ports. It keeps access governed through folder allowlists, a command whitelist, and rate limits so clients gain a large operating surface without unrestricted shell access.
      MIT