omni
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| OMNI_CDP_URL | No | The loopback CDP URL of an already-running Chrome/Chromium instance that Omni attaches to. Chrome must already expose loopback CDP; Omni never launches a browser/profile. Default is port 17082. CLI accepts --cdp-url, MCP reads OMNI_CDP_URL. | http://127.0.0.1:17082 |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| experimental | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| list_tabsB | List stable page-target IDs, titles and URLs without selecting/focusing a tab. |
| gotoA | Navigate an explicit tab or create one only with new_tab=true. No hidden observe. |
| observeC | Clean content, links and controls with exact refs; continue all streams using next_cursor. No field values. |
| actB | Native click/fill/press/select on an exact observed ref. No replay or hidden observe. select is native HTML only. |
| downloadB | Cookie-scoped HTTP GET streaming; atomic output, no Save dialog. No POST/blob/challenge fallback. |
| screenshotB | Return actual PNG ImageContent and metadata; optional safe output file. No focus/resize/scroll. Max 6M pixels, 8192px/side, 4 MiB. |
| scrollC | One main-document native wheel input. CSS pixels; reports before/after movement, not lazy-load completeness. Inner panes unsupported. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 7 tools
Each tool targets a distinct browser primitive: list_tabs (enumerate), goto (navigate), observe (read state), act (mutate state), download (fetch file), screenshot (capture image), scroll (viewport movement). The descriptions explicitly delimit boundaries (e.g. observe returns refs but no field values, act works only on observed refs), leaving no real overlap. observe vs screenshot are the closest pair but are clearly separated as DOM-content vs PNG image.
Six of seven names are clean single-verb actions (goto, observe, act, download, screenshot, scroll), giving a predictable action-oriented convention. list_tabs is the lone verb_noun outlier, a minor deviation rather than a different style. Overall readable and consistent.
Seven tools is well-scoped for a browser automation server, covering the essential perceive/act/navigate/capture primitives. Each tool earns its place with no redundant or filler entries. Falls squarely in the ideal 3-15 range.
Core browser lifecycle is covered: tab enumeration, navigation, observation, interaction, scrolling, screenshots, and downloads. Minor gaps exist, such as no explicit tab-close or history back/forward operation and no dedicated wait-for-load primitive, but agents can work around these via goto and repeated observe. The surface is solid for the stated purpose.