lmstudio-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| LMSTUDIO_BASE_URL | No | Base URL for LM Studio server. Must use http:// and a loopback hostname. | http://127.0.0.1:1234 |
| LMSTUDIO_LMS_PATH | No | `lms` executable name or absolute path. | lms |
| LMSTUDIO_API_TOKEN | No | Optional local bearer token; never persisted by this project. | |
| LMSTUDIO_DEFAULT_MODEL | No | Default model identifier for tests. | |
| LMSTUDIO_REQUEST_TIMEOUT_MS | No | HTTP/CLI timeout, bounded to 250–300000 ms. | 120000 |
| LMSTUDIO_MCP_ALLOW_MUTATIONS | No | Enables the two explicit load/unload tools. | false |
| LMSTUDIO_LOG_CAPTURE_MAX_SECONDS | No | Log stream duration, bounded to 1–120 seconds. | 30 |
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 |
|---|---|
| lmstudio_statusA | Return a read-only local LM Studio health summary, including API and model counts. |
| lmstudio_diagnoseB | Run a high-level, non-mutating diagnostic suite against local LM Studio. |
| lmstudio_validate_mcpA | Read and validate LM Studio's MCP JSON without changing it; secret-like values are redacted. |
| lmstudio_list_modelsA | List locally downloaded models and discoverable metadata. Read-only. |
| lmstudio_list_loaded_modelsA | List model instances currently loaded in memory. Read-only. |
| lmstudio_inspect_modelA | Return all discoverable local metadata and distinguish static capability from runtime configuration. |
| lmstudio_test_inferenceA | Run one tiny deterministic inference against an explicitly selected local endpoint. Never falls back to cloud. |
| lmstudio_test_endpointsC | Classify native chat, OpenAI chat completions, and Responses API as passed, unsupported, or failed. |
| lmstudio_test_tool_callingB | Run a real deterministic local tool-call round trip with argument validation, fixture execution, result delivery, and continuation verification. |
| lmstudio_test_streaming_tool_callingA | Verify a real streamed Chat Completions tool-call round trip, including fragmented delta reassembly, strict argument validation, fixture execution, streamed continuation, progress, and cancellation. |
| lmstudio_test_structured_outputA | Test JSON Schema enforcement independently from JSON mode and prompt-only JSON, then validate output with Ajv. |
| lmstudio_capabilitiesD | Separate DECLARED metadata, passively DETECTED support, and live VERIFIED capability evidence recorded in this MCP process. |
| lmstudio_security_auditA | Read-only audit of loopback binding, LAN exposure, CORS, authentication, sensitive logging, and remote destinations. |
| lmstudio_export_capability_snapshotB | Return a deterministic-schema, secret-free capability snapshot plus canonical JSON and its SHA-256 digest. This tool does not write files or rerun capability tests. |
| lmstudio_benchmarkB | Run at most five small streaming requests and report only measurements actually available. |
| lmstudio_capture_logsC | Capture a bounded lms log stream, redact secrets, and always terminate the child process. |
| lmstudio_load_modelA | MUTATING and disabled by default. Validates a downloaded model and context before invoking lms load; never downloads. |
| lmstudio_unload_modelA | MUTATING and disabled by default. Requires one explicit loaded identifier and never unloads all models. |
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 18 tools
Each tool has a clearly distinct purpose: status, diagnostics, validation, model listing, inspection, various tests, security audit, snapshot export, log capture, and load/unload actions. Even the testing tools are separated by specific API features (inference, endpoints, tool calling, streaming, structured output), leaving no real ambiguity.
All tools share the lmstudio_ prefix and use snake_case, but there is a mix of bare nouns (status, capabilities, benchmark) and verb_noun compounds (list_models, load_model, test_inference). This is mostly consistent and readable, but not perfectly uniform.
With 18 tools, the server is slightly above the typical well-scoped range, but the breadth is justified by its comprehensive diagnostic and testing mission. The tools cover distinct aspects of local LM Studio management and verification, so the count feels reasonable rather than bloated.
The toolset covers the full lifecycle of interacting with a local LM Studio: listing and inspecting models, loading/unloading, testing core API paths (inference, tool calling, structured output, streaming), health/diagnostics, security auditing, configuration validation, and log capture. No critical dead ends or missing operations are apparent for the stated purpose.