MCP server for controlling Reaper DAW (Digital Audio Workstation), enabling music production tasks like track creation, MIDI editing, VST plugin management, and sample library integration
ReaperMCPServer has significant structural quality gaps. While 15 tools are defined with visible schemas and descriptions, the descriptions are uniformly generic and lack the specificity required for LLM tool selection. Most descriptions are under 100 characters and do not explain WHEN to use each tool, WHAT it returns, or dependencies between operations. Parameter descriptions are minimal (e.g., 'Index of the track' for track_index appears in 7+ tools without context about valid ranges or prerequisites). No input validation, error handling guidance, or output schemas are documented. Tools like add_notes_to_the_active_take_of_item accept a generic 'list' type for notes with only a high-level description, forcing LLMs to infer the note object structure. The reaper_project_state tool returns str(state) instead of a structured dict, violating the response-shaper pattern. No pagination for list operations. Destructive tools (delete_track, delete_item, delete_fx_on_track) have no confirmation/dry-run pattern. Security: the instructions embed a hardcoded sample library path and contain narrative guidance rather than structured tool behavior. Overall, the tools map to a coherent domain (DAW control) but lack the rigor and LLM-optimization expected of production tools.
Adds a new item to the given track in the current reaper project
Adds given list of notes [{start, end, length, channel, pitch, velocity}, ...] to the active take of given item in the current reaper project. if no take exists, it will be created
Adds given vst plugin to the given track in the current reaper project
Creates a new track and inserts it at the given index in the current reaper project
Deletes the given fx on the given track in the current reaper project
Deletes the given item in the current reaper project
reaper_project_state returns str(state) instead of structured dict, violating response-shaper pattern. LLMs cannot parse string representations reliably and waste tokens on string parsing.
Destructive tools (delete_track, delete_item, delete_fx_on_track) lack confirmation or dry-run patterns. Agents cannot safely explore whether a delete is the right action before executing irreversible operations.
Parameter descriptions are generic and lack actionable constraints. 'Index of the track' appears 7+ times without explaining valid ranges, whether 0-based, or what happens if the index is out of bounds. LLMs cannot validate inputs without explicit bounds.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 48 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 28 | - | v1 |
Deletes the given track in the current reaper project
Explore the sample library and return a hierarchical list of samples/folders at the given path. provide the path relative to the root path to sample library
Returns state of the given fx on the given track in the current reaper project
Returns list of fxs on the given track in the current reaper project
Returns list of vst plugins
Inserts the given sample into the given track in the current reaper project. provide the path relative to the root path to sample library
Get reaper project state
Sets params of the given fx on the given track in the current reaper project, params is a list of dictionaries with name and value
Sets the bpm of the current reaper project
add_notes_to_the_active_take_of_item accepts 'notes' as a generic list type with only a high-level description 'List of note objects with start, end, pitch, velocity, channel properties'. The input schema does not specify the structure of each note object (no item schema for array elements), forcing LLMs to infer structure from the docstring or trial-and-error.
Tool descriptions are uniformly under 100 characters and do not explain WHEN to use each tool vs alternatives, WHAT it returns, or dependencies. E.g., 'Returns list of vst plugins' does not explain whether this lists available plugins or installed plugins, or how the result is used downstream.
No error handling guidance. If delete_track fails because the track is locked or contains active media, the tool provides no recovery hint (e.g., 'Try removing all items first' or 'This is a master track and cannot be deleted'). LLMs receive raw errors with no actionable next steps.
set_fx_params accepts params as an 'object' type with description 'Dictionary of parameters with name and value keys', but the actual parameter structure is under-specified. The implementation accesses param['name'] and param['value'], implying the schema should specify that params is an array of objects with 'name' and 'value' fields, not a single object.
insert_sample_into_track parameter 'start_time' is typed as array of numbers but has no bounds, units, or format guidance. LLMs cannot determine whether values are in seconds, milliseconds, measures, or beats without explicit specification.
No output schemas documented for any tool. LLMs must infer return types from descriptions alone. For example, get_list_of_vst_plugins returns a list, but the schema for each list element is not specified. This forces LLMs to guess field names and types.
explore_sample_library and insert_sample_into_track require paths relative to a hardcoded sample library root. Instructions embed this constraint as narrative ('provide the path relative to the root path to sample library') rather than as a parameter constraint (minLength, pattern, or description with explicit rules). This is fragile and non-machine-parseable.