local-agent-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| AGENT_DEBUG | No | When true, full prompts are written to the stderr debug log. | false |
| AGENT_ALLOW_WRITE | No | When true, permits codex_run workspace_write and opencode_run auto_approve. | false |
| AGENT_ALLOWED_ROOTS | Yes | Comma-separated absolute directories agents may access. | |
| AGENT_MAX_CONCURRENCY | No | Max simultaneous agent runs. | 3 |
| AGENT_MAX_OUTPUT_BYTES | No | Max combined stdout+stderr bytes captured per run. | 5000000 |
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": true
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| agent_healthA | Report the local agent environment: Node version, Git availability, Codex/OpenCode install status and versions, allowed working directories, whether writes are permitted, and current concurrency usage. |
| codex_runA | Run the locally installed, logged-in Codex CLI non-interactively in a sandboxed working directory. Defaults to read-only. Use workspace_write only when writes are enabled server-side. Returns the final agent message plus command/file-change/error summaries. |
| opencode_runA | Run the locally installed, logged-in OpenCode CLI non-interactively in a sandboxed working directory. auto_approve enables side-effecting actions and requires server-side write permission. Supports session continuation and agent/model selection. |
| agent_compareA | Run Codex and OpenCode against the same prompt in READ-ONLY mode and return both results verbatim. Does not judge which is better; the caller synthesizes the conclusion. One agent failing does not suppress the other. |
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 4 tools
Each tool has a distinct purpose: agent_health reports environment, codex_run runs Codex, opencode_run runs OpenCode, agent_compares both. No overlap or ambiguity.
All tools use snake_case with descriptive names, but there's a mix: agent_health is noun-based while codex_run, opencode_run, and agent_compare are verb-focused. Still predictable and clear.
4 tools is well-scoped for the domain of running and comparing local AI agents. Each tool earns its place without redundancy or bloat.
Covers health check, execution of both agents, and comparison. Minor gaps like lacking a tool to stop running agents or manage sessions, but core workflows are covered.