Skip to main content
Glama
ru-aish

Terminal GPT Experimental MCP

by ru-aish

Server Configuration

Describes the environment variables required to run the server.

NameRequiredDescriptionDefault
MCP_HOSTNoHTTP listener address127.0.0.1
MCP_PORTNoHTTP listener port8000
NGROK_URLNoReserved ngrok URL/domainrandom URL
MCP_LOG_DIRNoRuntime logs directory.run/logs through start.sh
MCP_GPT_HOMENoChanges the global GPT directory~/.GPT
MCP_VENV_DIRNoOverride the virtual-environment directory.venv
MCP_TRANSPORTNoTransport mode: stdio, sse, or streamable-httpstreamable-http
MCP_WORKSPACENoDefault terminal workspace~/mcp_workspace
MCP_AUTO_SETUPNoRun setup.sh automatically when the Python runtime is missing or incomplete1
MCP_SKIP_NGROKNoSet 1 for local-only mode0
MCP_BEARER_TOKENNoRequired bearer token for HTTP requests
MCP_CHAT_AGENT_DBNoShared SQLite ledger for watchdog task generations plus agent registry, tasks, commands, events, subscriptions, and cursors~/.GPT/chat-agent-orchestrator.db
MCP_DASHBOARD_TOKENNoRequire the dashboard login form; when unset, the dashboard inherits the reachability of the HTTP server
MCP_CHAT_WATCHDOG_CDPNoCodex desktop CDP endpointhttp://127.0.0.1:9222
MCP_AUTO_INSTALL_NGROKNoLet start.sh invoke store.sh --with-ngrok when ngrok is missing0
MCP_CHAT_AGENT_ENABLEDNoEnable the restart-safe parent/child orchestration sync loop for SSE/Streamable HTTP1
MCP_PROXY_IDLE_TIMEOUTNoDownstream MCP idle timeout in seconds1800
MCP_CHAT_WATCHDOG_MODELNoPreferred live model slug; otherwise reuse the thread model or live default
MCP_CHAT_WATCHDOG_QUEUENoEditable one-active-task-URL-per-line input queue; task state is reconciled into the durable ledger~/.GPT/chat-watchdog/threads.txt
MCP_CHAT_WATCHDOG_STATENoCompatibility mirror and one-time import source for watchdog task state~/.GPT/chat-watchdog/state.json
MCP_CHAT_AGENT_TAIL_LINESNoDefault recent visible/tool-summary lines60
MCP_CHAT_WATCHDOG_ADAPTERNoUse the Codex desktop app's internal normal-chat client; no UI fallback is enabledcodex-internal
MCP_CHAT_WATCHDOG_DRY_RUNNoInspect and report without sending0
MCP_CHAT_WATCHDOG_ENABLEDNoEnable the five-minute ChatGPT task watchdog for SSE/Streamable HTTP1
MCP_GOAL_REMINDER_SECONDSNoSeconds before goal reminder is re-injected (15 minutes by default)900
MCP_WATCH_IMAGE_MAX_BYTESNoMaximum bytes for watch_image (20 MiB before base64 expansion)20971520
NGROK_TRAFFIC_POLICY_FILENoOptional ngrok Traffic Policy file
MCP_CHAT_WATCHDOG_PROJECTSNoSelected Project allowlist plus baselined/seen conversation IDs for new-thread discovery~/.GPT/chat-watchdog/projects.json
MCP_CHAT_AGENT_SYNC_SECONDSNoBackground orchestration sync interval15
MCP_CHAT_WATCHDOG_COMPLETEDNoAppend-only task completion history~/.GPT/chat-watchdog/completed.jsonl
CHAT_GATEWAY_THINKING_EFFORTNoReasoning effort attached to every durable agent create/continue requestfalls back to watchdog effort
MCP_CHAT_AGENT_STALE_SECONDSNoUnchanged canonical-node age before an active/waiting child is marked stale and its parent is notified600
MCP_CHAT_AGENT_TAIL_MAX_CHARSNoMaximum recent-tail text size12000
MCP_CHAT_WATCHDOG_APP_COMMANDNoInstalled Codex desktop launch command/usr/bin/codex-desktop
MCP_CHAT_WATCHDOG_REQUIRE_HIGHNoRefuse to silently lower the requested reasoning level1
MCP_CHAT_WATCHDOG_RETRY_SECONDSNoDuplicate-send cooldown300
MCP_ALLOW_UNAUTHENTICATED_PUBLICNoDangerous public-tunnel override0
MCP_CHAT_AGENT_CONTEXT_MAX_CHARSNoMaximum serialized compact execution context12000
MCP_CHAT_WATCHDOG_AUTO_START_APPNoStart the desktop host when offline, or reopen its primary renderer when CDP is up but the internal client is not ready1
MCP_CHAT_AGENT_CONTEXT_MAX_EVENTSNoMaximum compact execution events returned or journaled per delta60
MCP_CHAT_WATCHDOG_THINKING_EFFORTNoRequested live reasoning effortextended
MCP_CHAT_WATCHDOG_INTERVAL_SECONDSNoScheduled scan interval300
CHAT_GATEWAY_REQUIRE_HIGH_REASONINGNoAccept only high or extended effort for durable agent requestsfalls back to watchdog requirement
MCP_CHAT_AGENT_MAX_CONTINUE_ATTEMPTSNoHard ceiling for automatic child-task continuation attempts20
MCP_CHAT_WATCHDOG_MAX_CONTINUE_ATTEMPTSNoHard per-task continuation-attempt ceiling20
MCP_CHAT_WATCHDOG_RUNNING_RESTART_SCANSNoStop and continue after this many consecutive scans observe the same generation working, even if it makes progress2
MCP_CHAT_WATCHDOG_STREAM_TIMEOUT_SECONDSNoMaximum time to wait for a watchdog-owned completion stream3600
MCP_CHAT_WATCHDOG_RUNNING_RESTART_SECONDSNoStop and continue a watched generation still working after twenty minutes1200
MCP_CHAT_WATCHDOG_INTERNAL_TIMEOUT_SECONDSNoTimeout for CDP attachment and internal reads10
MCP_CHAT_WATCHDOG_PRE_SEND_CONFIRM_SECONDSNoDelay before the canonical pre-send refetch1.25
MCP_CHAT_WATCHDOG_STALE_GENERATION_SECONDSNoAge after which unchanged in-progress canonical state is reported as stale; running recovery applies at the limits below600
MCP_CHAT_WATCHDOG_APP_START_TIMEOUT_SECONDSNoMaximum time to wait for a usable primary renderer after starting or reopening the desktop host30

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