Lwa MCP
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| LWA_MCP_START_DASHBOARD | No | Set to '1' to launch the dashboard from the same MCP server process. If unset or empty, the dashboard is not started by the MCP server, avoiding handshake delays. |
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 |
|---|---|
| route_taskC | Prepare and lock the selected provider/model before Working may begin. |
| prepare_taskC | Complete silent preflight. Do not show Working until working_may_begin is true. |
| approve_preflightB | Approve a locked preflight. This does not begin provider work. |
| run_prepared_taskC | Begin provider work for one approved plan. Never routes or switches models. |
| set_tier_guidanceC | Enable compact post-response tier guidance: local or codex_request. |
| smart_completeB | Prepare a locked task. Despite the legacy name, this never begins provider work. |
| confirm_and_runC | Deprecated compatibility alias: approve preflight only, then call run_prepared_task. |
| quick_responseC | Prepare a low-latency auxiliary response; execution requires run_prepared_task. |
| answer_queryC | Prepare a routed query; execution requires run_prepared_task. |
| verify_workB | Prepare a verification route; execution requires run_prepared_task. |
| generate_imageD | Prepare and lock an image-generation route before execution. |
| generate_videoC | Prepare and lock an official video-generation route before execution. |
| compress_conversationC | Prepare conversation compression while preserving required context. |
| optimize_promptC | Prepare token optimization while preserving binding requirements and literals. |
| edit_documentC | Prepare conservative document editing; execution requires run_prepared_task. |
| write_sitrepC | Prepare an evidence-bound SITREP route before Working begins. |
| plan_workD | Prepare project planning with its prompt, budget, and model locked. |
| build_consensusD | Prepare all eligible free-model voters and a synthesis route before Working. |
| set_switch_confirmationB | Set model-switch prompting: always_ask, paid_only, or automatic. |
| refresh_model_catalogB | Refresh live model catalogs and provider health without exposing API keys. |
| refresh_provider_quotasB | Fetch official remote balance/quota data for providers with configured management endpoints. |
| router_statusA | Return local usage, latest quota headers, provider health, and consent mode. |
| pipeline_statusB | Check pipeline readiness without sending a user prompt. The default check is local-only. |
| mark_pipelineB | Record an operator pipeline mark: ready, degraded, blocked, or maintenance. |
| invalidate_pipeline_probe_cacheB | Invalidate cached zero-payload pipeline probes without provider I/O. |
| model_tiersB | List model tiers sorted by maximum declared free-plan token usage. |
| syntax_contract_resourceA | Return canonical task, quality, and prompt-variable syntax. |
| list_library_toolsB | List persistent reusable tools. Status may be active, draft, disabled, or archived. |
| search_tool_libraryB | Search reusable cross-project tools by task, words, tags, and trigger descriptions. |
| suggest_library_toolsB | Find active library tools that are pragmatic and relevant to the current work. |
| read_library_toolA | Read a reusable tool's manifest, README, path, origin, evidence, and usage instructions. |
| run_library_toolC | Prepare an active prompt recipe; approved script tools remain separately executable. |
| create_library_toolC | Create a documented reusable prompt tool with a manifest, README, and launcher script. |
| register_script_toolA | Catalog a script as a draft tool. It cannot execute until separately reviewed and enabled. |
| approve_library_toolB | Mark a reviewed tool active and approved. Script execution still obeys global configuration. |
| disable_library_toolB | Disable a reusable tool without deleting its evidence, README, or version history. |
| observe_workflowB | Record one workflow occurrence; repeated successful patterns may create a library tool. |
| analyze_repeated_workflowsC | Review the repetition ledger and scaffold any eligible reusable tools not yet created. |
| rebuild_tool_catalogB | Regenerate the human-readable CATALOG.md for cross-conversation tool discovery. |
| library_statusA | Return persistent tool-library paths, counts, and recent detected patterns. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
| preflight_guidance | Provide a concise host-side checklist for strict Lwa task execution. |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| status_resource | Read-only operational status without credentials or raw prompts. |
| catalog_resource | Read-only provider/model catalog with fallback/live source metadata. |
| model_tiers_resource | Configured model tiers ranked by declared free-plan token capacity. |
TDQS
Scored across 40 tools
Several tools occupy the same prepare/lock/run lifecycle—prepare_task, route_task, smart_complete, answer_query, quick_response, verify_work, plan_work, and confirm_and_run—so an agent can easily select the wrong one. The deprecated confirm_and_run alias and legacy smart_complete name add further ambiguity despite descriptions attempting to differentiate them.
Most tools use snake_case verb_noun names, but there are noun-only status tools (pipeline_status, router_status, model_tiers, library_status), a resource noun (syntax_contract_resource), and legacy/multi-verb names (smart_complete, confirm_and_run). The convention is readable but not consistently applied.
40 tools is far beyond the comfortable MCP range, and many are granular variants of the same prepare-before-run flow. The broad domain explains some of the count, but the set would be tighter if the shared execution path were factored into fewer tools.
The lifecycle is broadly covered: task preparation/approval/execution, library tool management, provider refresh, and workflow analysis. However, library tools lack update/delete operations and several prepared actions depend on a generic run_prepared_task, leaving some gaps in the apparent CRUD surface.