voicehook MCP Server
by voicehook-aiio.github.voicehook-ai/voicehookv1.0.0
Talk with your agents. A voice call in the browser that your agent can start or join.
context tax
~4,217 tok/turn10 tools
security
97/100
cold start
hosted
freshness
Active1d ago
Our reading of voicehook, from the data
voicehook is a browser automation MCP server from voicehook-ai, listed in the official MCP Registry.
Its 10 tools add about 4,217 tokens to every model turn, heavier than the browser automation median of 1,697 tokens across 116 measured servers (measured 2026-10-19).
Maintenance is active: the last public signal was in the last day (2026-10-18). For comparison, 67% of browser automation servers are active or maintained.
With 0 GitHub stars it ranks #560 of 1,199 browser automation servers we index.
It runs as a hosted endpoint, so there is nothing to install.
Install voicehook MCP server
Install in Claude Code
claude mcp add --transport http voicehook-agent https://voicehook.ai/mcp?utm_source=mcpregistry&utm_campaign=d11-registryInstall in Cursor
{
"mcpServers": {
"voicehook-agent": {
"url": "https://voicehook.ai/mcp?utm_source=mcpregistry&utm_campaign=d11-registry"
}
}
}Install in Claude Desktop
{
"mcpServers": {
"voicehook-agent": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://voicehook.ai/mcp?utm_source=mcpregistry&utm_campaign=d11-registry"
]
}
}
}Or add it as a custom connector under Settings → Connectors. The config above bridges via mcp-remote.
Install in VS Code
{
"servers": {
"voicehook-agent": {
"type": "http",
"url": "https://voicehook.ai/mcp?utm_source=mcpregistry&utm_campaign=d11-registry"
}
}
}Install in Windsurf
{
"mcpServers": {
"voicehook-agent": {
"serverUrl": "https://voicehook.ai/mcp?utm_source=mcpregistry&utm_campaign=d11-registry"
}
}
}Remote endpoints
- streamable-http
https://voicehook.ai/mcp?utm_source=mcpregistry&utm_campaign=d11-registry
Context tax
Connecting this server adds ~4,217 tokens of tool definitions to every model turn (moderate, 10 tools). Measured 2026-10-19 via its hosted endpoint; o200k tokenizer, worst case for clients that load all tools up front. How we measure
| Model | Per 1,000 turns | With prompt caching |
|---|---|---|
| Claude Opus 5.5 | $16.87 | $0.843 |
| Claude Sonnet 5.5 | $8.43 | $0.843 |
| Claude Haiku 5.5 | $0.422 | — |
Input prices as of 2026-10-06. Excludes your own prompt and output.
Tools (10)
status833 tokReplace the agent's status board on the call page with this one (never spoken; the voicebot answers the person's questions about progress from it). doing: current step and ETA (max 400 chars). open / done: max 10 items each, 200 chars per item. faq: up to 6 question/answer pairs the voicebot can answer directly. emoji (optional): An emoji shown on the agent's chip in the call, from the listed set. Empty string removes it, left out keeps it. Whole board max 2000 chars. Returns ok and humans_present.
join682 tokConnect to a voicehook.ai voice call as the agent behind its voicebot. invite_url: the full invite link including ?invite= (from the person or from create_call). label: the text on the agent's chip in the call, max 22 characters (longer is rejected). model: the model id the agent runs on (internal, not shown). greet (default true): the voicebot speaks a short greeting with the label in the call language right after the join (or when the person is back, if they are out). Returns `session` (the handle for every other tool, valid until the call ends, after 10 min without next or say, or after lea
send_data580 tokSend JSON, text or a file to other agents in this call (silent data channel: never spoken, never shown to the human, never to the voicebot; the human only sees one activity line 'Data to B: name (size)'). to: the other agent's label (as on its chip), its identity, or "*" for all other agents; unknown -> 404 with the labels in the call. kind json: payload any JSON; text: payload a string; file: payload = base64 of the file plus name and mime, or file_id of a file uploaded with the REST endpoint POST /api/bridge/upload (files over about 11 KB, up to 10 MB). The whole packet is max 15000 bytes in
next461 tokWait (up to `timeout` seconds, max 50) for the next event in the call and return it. Event `type`: user (`text` = what the person said; after a pause also `missed` = earlier turns, oldest first, and `away_s`), revise (queued says were replaced; reason user_stop = the person said stop), status_request (the person asked what the agent is doing), decision (answer to a decide question: id, number, verdict yes|no|option|other, option, text, by), timeout (nothing happened; reconnect true = server restart), human_left / human_rejoined (the person dropped out / is back; greeted true = the join greetin
decide391 tokPut a question for the person on the call's Decision Board (shown as #number with buttons; the person can also answer by voice, also freely). Without options it is a yes/no question (buttons Yes/No); options: 2 to 3, max 40 characters each, recommend = index of the recommended option. question: max 160 characters. Longer or more: 422, never cut. Returns id and number. The answer arrives later as a `next` event {type: decision, id, number, verdict: yes|no|option|other, option?, text? (their own answer), by}. Change an open one with id plus new fields, remove it with id and delete true.
show378 tokPut a card on the person's call page (never spoken): text, code, a link or an image. kind: text (default), code (monospace with copy button), link (exactly one http(s) URL, shown as a tappable link), image (`image` = SVG text or base64 PNG/JPEG/WebP). A new card appears as a small preview top left; panel cards opens the large view, board, log or decisions open those panels, none closes them. Limits: text 4000 chars, title 80, SVG 100 KB, raster 1024 KB.
create_call339 tokStart a new voicehook.ai voice call. `lang` = the language of the person being called (de or en, default de). Returns `human_url` (the link for the person; whoever opens it starts the call, on their own free allowance or credit), `share_text` (a ready message with human_url and the hint that the link needs a normal browser, because an app's built-in browser blocks the microphone), `operator_invite_url` (the agent's own link for `join`), `room` and `expires_in` (seconds until an unopened link expires). Errors: 429 = too many new calls right now.
say183 tokThe voicebot speaks `text` aloud in the call (verbatim in normal mode; in a live room the voice rephrases it and the answer has verbatim false, mode live_paraphrase). mode: append (default, queued after earlier says), overwrite (replaces this agent's says that have not started yet), revise (like overwrite, and next then returns a revise event). Returns ok and seq.
leave109 tokLeave the call (optionally speaking `say` first, e.g. a short goodbye). The session handle is invalid afterwards.
activity96 tokAppend one line (3-8 words) to the activity feed on the call page, shown as a timestamped line. Never spoken.
Security signals
97/100 — no known issues found as of 2026-10-19, from the checks below. This is not a guarantee of safety; review what a server can do before connecting it.
Deductions
- No security policy (SECURITY.md)-3
Checks
- ✓Repository hygiene (archived, activity, license, security policy)
- ✓Tool descriptions (prompt-injection and hidden-text heuristics)
- ○Package metadata (install scripts, provenance) — pending
- ○Source scan (secrets, CVEs, risky code) by vectra-guard — pending
Scan it yourself before installing with vectra-guard, the open-source agent security CLI. Scoring rules
Freshness
Active — last maintenance signal 1d ago. The newest of the signals below sets the band.
Last commit (default branch)
2026-10-18 · 1d ago · GitHub
Latest release
2026-10-12 · 7d ago · GitHub · cli-v0.14.3
Package published
no data · npm/PyPI
Registry entry updated
2026-10-17 · 2d ago · official registry · v1.0.0
How it compares in Browser Automation
Against the most-starred browser automation servers we index. Token cost is per model turn, measured from each server's live tool list.
| Server | Stars | Tools | Tokens / turn | Freshness | Runs |
|---|---|---|---|---|---|
| voicehook (this) | 0 | 10 | 4,217 | Active | hosted |
| Browser Use | 118,444 | — | not measured | Active | local |
| Scrapling | 88,517 | — | not measured | Active | local |
| Browser | 39,318 | — | not measured | Active | local + hosted |
| Playwright | 38,349 | — | not measured | Active | local |
| Browser | 36,690 | — | not measured | Active | — |
FAQ
›How do I install the voicehook MCP server in Claude Code?
Run: claude mcp add --transport http voicehook-agent https://voicehook.ai/mcp?utm_source=mcpregistry&utm_campaign=d11-registry. For Cursor, VS Code, Claude Desktop and Windsurf, use the install tabs above.
›Does voicehook require an API key?
No required environment variables are declared in its published metadata.
›Can I use voicehook as a remote (hosted) MCP server?
Yes. It is a hosted MCP server; connect to its URL with any client that supports remote MCP.
›Is voicehook in the official MCP registry?
Yes, as io.github.voicehook-ai/voicehook.
›How many tokens does voicehook add to my context?
Its tool definitions total about 4,217 tokens, resent on every turn.
Sources and method
- Official MCP Registry entry
- updated 2026-10-17
- GitHub repository
- fetched 2026-10-19
- Tool list (tools/list), measured by us
- 2026-10-19
- Security signals, computed by us
- 2026-10-19
- Last checked against all sources
- 2026-10-19
The description, packages and links come from the maintainer's own published metadata. Token cost, security signals, freshness and the reading above are MCPSkillHub's own measurements; see the methodology. Editorial standards are owned by Ramachandra Vikas Chamarthi, Founder & AI HPC Architect. Spot an error? Request a correction or claim the listing.
Alternatives to voicehook
Other browser automation MCP servers.