LiuHe
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| MALONG_PORT | No | TCP port for the parse daemon (Windows). | 31001 |
| MALONG_SOCKET | No | Unix socket path to the parse daemon (Linux / macOS). | /tmp/malong-parse-$(id -u).sock |
| MALONG_PARSE_BIN | No | Binary used when the client auto-starts the daemon. | npm platform pkg / ~/.local/bin |
| MALONG_STATE_DIR | No | Where usage / feedback / edit-stats files are written. Override to redirect state (tests, sandboxed hosts). | ~/.config/malong |
| MALONG_PARSE_MODE | No | Parse transport. Only 'rust-service' is supported. | rust-service |
| MALONG_WS_GC_DAYS | No | Days a workspace index cache may sit untouched before health cleanup prunes it; 0 disables. | 14 |
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 | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| code_auditA | Read-only review pass (code review, security, dead code…) on files you touched or any scope — reports findings via checks, changes nothing. Zero findings is not proof. |
| code_batchA | Apply edits to N files in ONE atomic call — all or nothing; single edit → code_patch. New: create:true + content + locator.file_path. dry_run default preview. |
| code_checkA | Run test/typecheck/lint after editing — records commands + results as task evidence; better than raw shell. Previews exact commands first; async:true for long suites. |
| code_controlC | Manage task lifecycle: cancel a running task, resume a dropped one (returns a self-contained brief), reopen a finished one, check a background job. |
| code_maintainB | Clean up imports in files you touched (missing/unused). plan = preview by default; dead code is never removed automatically, only flagged. |
| code_patchA | Edit any text file (code, md, config) — a reversible alternative to Edit/Write: patch text you already read (pass its read_ref), or create:true + content + locator.file_path. dry_run default: preview; false applies. |
| code_queryB | Look things up fast, no task needed: where is X, who calls it, what's in this file. mode: outline|symbols|body|map|search|impact|references|callers|callees|tests|validate|capabilities. Read-only. |
| code_readA | Read a file or one symbol before editing; returns content + a read_ref to pass back when editing. budget_hint=0 → full file. |
| code_statusA | See what a task changed so far: state, files, evidence. Omit task_id → list active tasks. |
| code_taskA | Start here for any change: describe the goal, get candidates + structure + impact in one call. Planning only — does not itself edit or check; then code_read → code_patch / code_batch → code_check. |
| code_worktreeA | Land a risky change safely: apply runs edits in a temp git worktree, your verify command gates it, and only on pass does the commit fast-forward into the CURRENT branch. |
| edit_transactionA | Multi-file edit with rollback/undo (begin->edit->commit); single-file edits don't need one. |
| invoke_capabilityA | Escape hatch: call a tool not listed above, by exact name; delegates once per call — no retries. Prefer listed tools; read-only unless the target itself writes. |
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 13 tools
The editing surface is crowded: code_patch, code_batch, edit_transaction, and code_worktree all mutate files, and the boundary between code_batch's atomic multi-file apply and edit_transaction's multi-file rollback/undo is not sharp. code_read and code_query also overlap on reading file contents, though descriptions provide some guidance.
All tools use snake_case and ten of thirteen share a code_ prefix; edit_transaction and invoke_capability are exceptions but remain readable. The mix of verb and noun suffixes (code_read vs code_task/code_worktree) is a minor inconsistency.
13 tools is within the well-scoped 3–15 range for a code-editing workflow. Each tool maps to a planning, reading, editing, verification, maintenance, or lifecycle concern.
The surface covers planning, querying, reading, single/multi-file edits, transactions, worktrees, checks, audits, maintenance, and task control. Minor gaps remain around explicit file deletion/rename and first-class diff/undo of applied patches.