Skip to main content
Glama

Server Configuration

Describes the environment variables required to run the server.

NameRequiredDescriptionDefault

No arguments

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

CapabilityDetails
tools
{}

Tools

Functions exposed to the LLM to take actions

NameDescription
zcode_statusB

Report the state of this MCP server and of ZCode itself. Read-only; starts no turn. "probe" is the diagnostic entry point (runtime version, protocol identity, session count).

zcode_sessionC

ZCode session lifecycle and per-session settings. Mutating actions re-read the session and fail if the observed value disagrees with the request.

zcode_chatB

Submit work to the ZCode agent and follow it to a terminal state. Success is only reported after a terminal turn event is observed; an accepted-but-unobserved command is reported degraded. Use tool_allowlist:["Read"] for a guaranteed non-mutating turn.

zcode_conversationA

Read a conversation: rows, messages, events, plans, usage. Read-only. Reads carry logEpoch and revision; on staleness the read is retried once, then reported degraded.

zcode_filesA

Inspect what a turn changed, preview a rewind, and move attachments. ZCode has NO editor document API: there is no "read the file as the editor sees it". File mutation goes through the agent's own Write/Edit tools, which is the only path that produces checkpoints and participates in rewind. rewind_apply is destructive and requires confirm:true.

zcode_settingsA

Read and change configuration. Protocol-backed actions take effect immediately; file-backed actions (set_desktop) are read by ZCode at startup and report that a restart is required. Provider mutations require ZCODE_MCP_ALLOW_PROVIDER_EDIT=1. Secrets are redacted in all output.

zcode_pluginsB

Enumerate and manage extensions. Enabling plugins consumes the model's tool budget: the provider rejects requests above roughly 89-94 registered tools with [1210] Invalid API parameter, so this warns when the count approaches the budget. install/update/uninstall require ZCODE_MCP_ALLOW_PLUGIN_INSTALL=1.

zcode_mcpA

Inspect and manage ZCode's MCP client surface. WARNING: "list" and "status" START the configured MCP servers as a side effect; the started instance ids are reported. Use "servers" to read configuration without starting anything. Config edits require ZCODE_MCP_ALLOW_MCP_CONFIG_EDIT=1 and a restart.

zcode_automationB

Manage scheduled agent runs. Creating an automation grants STANDING, UNATTENDED authority at the mode recorded at creation time. At most 20 automations are retained.

zcode_usageC

Token and activity analytics. Read-only.

zcode_modelsA

Discover models and providers, and select one. ACTION "catalog" is the decision surface: it lists what ZCode can talk to with context windows, modalities, reasoning levels and whether this server holds a credential for that provider — filterable, and deliberately NOT ranked, because choosing a model is the job of the caller. "available" reports what the runtime has wired up (a different question). "select" applies a choice at session, workspace or server scope; server scope affects only newly spawned runtimes in this process and does not persist.

zcode_approvalA

Answer the agent's approval, input and elicitation requests. Owning a runtime makes this server the runtime's only client, so unanswered requests block turns. The default policy is DENY. persist_rule writes a durable permission rule and requires ZCODE_MCP_ALLOW_PERSIST_RULES=1.

zcode_headlessB

One-shot headless run via the zcode CLI, requiring no protocol. Emits only flags verified to parse; unverified flags are reported as skipped. Needs a configured model provider.

zcode_protocolA

Escape hatch: raw ZCode Protocol access. Disabled by default (ZCODE_MCP_DISABLE_PROTOCOL=1 turns it off entirely) and restricted to an allowlist; mutating methods need ZCODE_MCP_PROTOCOL_ALLOW_MUTATIONS=1. Results are always marked unreliable because no read-back or schema guarantee is applied.

Prompts

Interactive templates invoked by user choice

NameDescription

No prompts

Resources

Contextual data attached and managed by the client

NameDescription

No resources

TDQS

A3.5/5.0

Scored across 14 tools

Disambiguation4/5

Each tool targets a distinct ZCode subsystem (status, session, chat, conversation, files, settings, plugins, MCP, automation, usage, models, approval, headless, protocol), limiting direct overlap. However, execution-related tools (zcode_chat, zcode_headless, zcode_automation, zcode_protocol) and read/analytics tools (zcode_conversation, zcode_usage) could be confused at a glance, so not entirely unambiguous.

Naming Consistency5/5

All 14 tools use the identical zcode_ snake_case prefix and noun-based naming pattern (zcode_status, zcode_session, etc.). No mixing of camelCase, verb styles, or other conventions.

Tool Count5/5

14 tools is within the well-scoped 3–15 range and matches the breadth of the ZCode platform. Each tool corresponds to a distinct subsystem rather than a trivial action, so the set is not bloated.

Completeness4/5

The surface covers core lifecycle and management areas: status, session, chat, conversation, files, settings, plugins, MCP, automation, usage, models, approval, headless, and raw protocol. Minor gaps remain—e.g., no explicit abort/cancel-turn operation and no direct file-read tool (an intentional architectural limitation)—so an agent may need workarounds for edge cases.

Maintenance

ActivityMaintained
ResponsivenessUnresponsive