Skip to main content
Glama

Server Configuration

Describes the environment variables required to run the server.

NameRequiredDescriptionDefault
ASR_HOSTNoLoopback ASR backend host; default localhostlocalhost
ASR_PORTNoLoopback ASR backend port; default 80008000
HF_TOKENNoHugging Face token with accepted pyannote model access; required for speaker diarization. Example: <HUGGING_FACE_TOKEN>
OCR_HOSTNoLoopback OCR backend host; default 127.0.0.1127.0.0.1
OCR_PORTNoLoopback OCR backend port; default 80028002
ASR_MODELNoExplicit local model or Hub ID; custom weights are outside the recorded 8 GB result
ASR_PROFILENo`default` uses Qwen3-ASR-1.7B; `8gb` selects the separately tested Qwen3-ASR-0.6B REST profiledefault
ASR_LOG_FILENoMCP startup error log path; default /tmp/qwen3-asr-server.log/tmp/qwen3-asr-server.log
OCR_JOB_ROOTNoDurable queue and artifact directory; defaults below $XDG_STATE_HOME/ocr/jobs$XDG_STATE_HOME/ocr/jobs
BRAVE_API_KEYNoBrave Search API credential; required for Brave Websearch. Example: <BRAVE_SEARCH_API_KEY>
OCR_MODEL_NAMENoReplace the OCR model ID/path; otherwise use MCP_TOOLS_MODEL_DIR/ocr and the Hugging Face cache/Hub
OCR_MODEL_ROOTNoReplace the OCR model local search root; otherwise use MCP_TOOLS_MODEL_DIR/ocr and the Hugging Face cache/Hub
OCR_USE_LAYOUTNoSet 0 to disable the isolated layout stage; default 11
ASR_IDLE_TIMEOUTNoSeconds before the ASR backend releases the GPU; default 300300
OCR_LAYOUT_MODELNoExplicit PP-DocLayoutV3 directory; otherwise use configured model roots or the PaddleX managed cache
MATHJAX_NODE_PATHNoOverride the pinned local MathJax Node entry point
VISION_LOCAL_PORTNoInteractive loopback port; default 80038003
MCP_TOOLS_MODEL_DIRNoShared OCR/Vision local model root; defaults to $XDG_CACHE_HOME/mcp-tools/models or ~/.cache/mcp-tools/models$XDG_CACHE_HOME/mcp-tools/models or ~/.cache/mcp-tools/models
OCR_JOB_TTL_SECONDSNoCompleted artifact retention; default 36003600
VISION_LOCAL_PROFILENo`default` uses the 9B interactive profile; `8gb` selects the bounded 4B interactive profiledefault
BROWSER_FETCH_TIMEOUTNoTimeout shown by browser_status; requests use timeout=30 unless explicitly overridden per call
BROWSER_FETCH_HEADLESSNoMode shown by browser_status; requests use headless=true unless explicitly overridden per call
VISION_LOCAL_MODEL_DIRNoRoot containing the Vision model directories used by all profiles; defaults to MCP_TOOLS_MODEL_DIR/visionMCP_TOOLS_MODEL_DIR/vision
BROWSER_FETCH_LOG_LEVELNoINFO or DEBUG; default INFOINFO
VISION_LOCAL_MODEL_PATHNoInteractive GGUF model path; required if the default model is absent. Example: /models/vision/model.gguf
BROWSER_FETCH_USER_AGENTNoOverride the browser user agent
VISION_LOCAL_MMPROJ_PATHNoInteractive vision projector path; required if the default projector is absent. Example: /models/vision/mmproj.gguf
VISION_LOCAL_SERVER_BINARYNollama.cpp server path; required if the default build is absent. Example: /opt/llama.cpp/bin/llama-server
BROWSER_FETCH_SCREENSHOT_DIRNoDefault PNG output directory; default /tmp/browser-fetch/tmp/browser-fetch

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

Server capabilities have not been inspected yet.

Tools

Functions exposed to the LLM to take actions

NameDescription

No tools

Prompts

Interactive templates invoked by user choice

NameDescription

No prompts

Resources

Contextual data attached and managed by the client

NameDescription

No resources