feneri.read_cliptool
Analyze a recorded clip: video, audio, text, or any combination. Returns a single multimodal state object.
args · path:string · modes:string[] · region?:string
NewFeneri MCP server · public beta
Plug Feneri into any agent.
A standards-compliant MCP server that exposes the emotional-intelligence engine as tools any agent can invoke: Claude Desktop, Cursor, Cody, and your own. One command. Six tools. Six seconds to setup.
$ npx -y @feneri/mcp install --host claude
Works with every MCP-compatible host
Install
Pick your host. Pick your shell.
One config edit and Feneri shows up as six tools your agent can reach for whenever it needs to read the person on the other side.
Claude Desktop.
Add Feneri to your Claude Desktop configuration. The app detects and loads it on next start.
One-click installOpens Claude Desktop and adds the server automatically
Add to Claude → Or edit config manually ·
~/Library/Application Support/Claude/claude_desktop_config.json {
"mcpServers": {
"feneri": {
"command": "npx",
"args": ["-y", "@feneri/mcp"],
"env": {
"FENERI_API_KEY": "fnr_••••"
}
}
}
}
Cursor.
Cursor reads MCP servers from settings. Add Feneri and it's available in every chat session.
Settings → MCP → Add new server
{
"name": "feneri",
"command": "npx -y @feneri/mcp",
"env": { "FENERI_API_KEY": "fnr_••••" }
}
Cody.
Use Sourcegraph Cody with Feneri as a custom tool provider. Works in VS Code and JetBrains.
.vscode/settings.json
"cody.experimental.mcp.servers": {
"feneri": {
"transport": "stdio",
"command": "npx",
"args": ["-y", "@feneri/mcp"]
}
}
Zed.
Zed's context-server feature picks up Feneri automatically once configured.
~/.config/zed/settings.json
"context_servers": {
"feneri": {
"command": { "path": "npx", "args": ["-y", "@feneri/mcp"] }
}
}
One-line install.
The installer figures out your host and writes the right config. Works with most popular MCP clients.
Terminal
# auto-detects Claude · Cursor · Cody · Zed
npx -y @feneri/mcp install
# or pick a host explicitly
npx -y @feneri/mcp install --host claude
Custom agent.
Feneri ships as a Node and Python package so any MCP-compatible runtime can spawn it as a subprocess.
Node SDK
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio";
import { FeneriServer } from "@feneri/mcp";
const server = new FeneriServer({ apiKey: process.env.FENERI_API_KEY });
await server.connect(new StdioServerTransport());
Python SDK
from mcp.server.stdio import stdio_server
from feneri_mcp import FeneriServer
server = FeneriServer(api_key=os.environ["FENERI_API_KEY"])
await stdio_server(server).run()
The toolset
Six tools your agent can reach for.
Standard MCP tools. Strongly typed. Streaming where it matters. Use one, use all six. The agent decides when to read the room.
feneri.read_streamstreaming
Open a live capture for the duration of an agent turn. Emits state updates every 33ms; closes when the turn ends.
args · source:enum · duration_s?:number · modes:string[]
feneri.summarize_sessiontool
Aggregate state across an entire session. Peaks, troughs, transitions, and a narrative summary the agent can repeat back.
args · session_id:string · format:'json' | 'text'
feneri.flag_distresstool
Boolean check: is distress above a threshold, sustained for N seconds? Designed for clinical and safety workflows.
args · threshold:number · sustained_ms:number · session_id?:string
feneri.compare_baselinestool
Diff two sessions: pre/post intervention, week-over-week, before/after a change. Returns deltas per signal.
args · session_a:string · session_b:string
feneri.subscribe_webhookresource
Tell Feneri to post back when state transitions cross a threshold. Stress spikes, attention dips, distress baselines shifting.
args · url:string · events:string[] · secret?:string
Quickstart From config
From config
to first call.
Once the server's installed, the agent does the work. Ask "How am I doing?" It picks up the camera and mic, calls read_stream, and brings back a real answer.
- Zero glue code · the agent picks the right tool
- Local capture · raw streams never leave your machine
- Streaming tools surface updates between agent turns
- Same API key works across every host
example · agent transcriptsimulated
// user prompts the agent
USER: How am I doing right now?
// agent reaches for the tool
AGENT: feneri.read_stream({
source: "webcam",
duration_s: 10,
modes: ["emotion", "attention", "stress"]
})
// engine streams back
RESULT: {
emotion: 0.62,
attention: 0.78,
stress: 0.41,
trend: "stress rising"
}
AGENT: You're focused but a little tense. Stress
ticked up over the last few seconds. Want to
pause for a minute?
Six tools. One install.
Free for sandbox keys. No card. The agent gets emotion, attention, and intent. You keep the credits for production.