claude_mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| CLAUDE_MCP_MODEL | No | Required Opus 5 model | claude-opus-5 |
| CLAUDE_MCP_STATE_DIR | No | Session registry and lock state directory | XDG state default |
| CLAUDE_CODE_OAUTH_TOKEN | No | Optional OAuth token for Claude Code authentication | |
| CLAUDE_MCP_ALLOWED_ROOTS | Yes | Allowed workspace roots (colon-separated on POSIX) | |
| CLAUDE_MCP_CLAUDE_BINARY | No | Claude Code executable | claude |
| CLAUDE_MCP_MAX_CONCURRENT | No | Concurrent distinct workspaces | 2 |
| CLAUDE_MCP_MAX_INPUT_CHARS | No | Maximum serialized task payload | 500000 |
| CLAUDE_MCP_TIMEOUT_SECONDS | No | Total completion timeout | 7200 |
| CLAUDE_MCP_MAX_OUTPUT_BYTES | No | Maximum Claude stdout | 16777216 |
| CLAUDE_MCP_MAX_STDERR_BYTES | No | Maximum captured stderr | 1048576 |
| CLAUDE_MCP_HEARTBEAT_SECONDS | No | MCP heartbeat interval | 30 |
| CLAUDE_MCP_KILL_GRACE_SECONDS | No | SIGTERM grace before SIGKILL | 10 |
| CLAUDE_MCP_MIN_CLAUDE_VERSION | No | Minimum supported CLI version | 2.1.224 |
| CLAUDE_MCP_REQUIRE_GIT_FOR_WRITES | No | Require Git for execution | true |
| CLAUDE_MCP_EXECUTE_PERMISSION_MODE | No | Claude execute permission mode | auto |
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
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
| experimental | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| opus_statusA | Check Claude Code version, subscription authentication, sandbox support, bridge configuration, and an optional workspace without invoking a model completion. |
| opus_consultA | Ask a workspace-aware Opus 5 session to plan, design, diagnose, or independently review without editing files. The 24-turn minimum is a completion ceiling, not a required amount of work; omit max_turns to use the reliable default. |
| opus_executeB | Delegate bounded implementation, fixes, refactoring, tests, or documentation to Opus 5 with direct write access; inspect the returned trusted workspace manifest. The 64-turn minimum reduces accidental early truncation of editing runs. |
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 3 tools
The three tools map onto a clear read-only / non-editing / editing split: opus_status inspects configuration, opus_consult plans and reviews without writes, and opus_execute delegates with write access. opus_consult and opus_execute could still be momentarily confused since they both invoke a model session, but the descriptions explicitly distinguish edit vs. no-edit behavior.
All three tools follow the same opus_<verb> pattern, with concise verbs (status, consult, execute) that consistently describe the action. No mixing of conventions or casing.
Three tools is minimal but each serves a distinct role in the delegate-to-Opus workflow, so nothing is redundant. It sits at the thin edge of the acceptable range, with no lifecycle tooling (cancel, resume, list sessions) to round it out.
Configuration check, planning/review, and execution are covered, but the surface has notable gaps: no way to list or resume sessions, cancel or abort a long run, or retrieve results of a previously delegated task. Agents must work around these lifecycle omissions.