MCP server for Neovim that spawns an embedded Neovim process and exposes tools for lifecycle management, interaction (commands, Lua, keystrokes), and state inspection.
nvim-mcp has well-structured tool definitions with complete input schemas and descriptions for all 13 tools. Naming follows verb_noun convention (nvim_start, nvim_execute, nvim_lua). However, descriptions are inconsistent in depth, some are terse (nvim_stop: 5 words), others detailed (nvim_start: 200+ chars). Parameter descriptions are present but often lack constraint details (e.g., nvim_execute's 'cmd' param has no format guidance; nvim_send_keys 'mode' param lists options but no explanation of when to use each). Output schemas are not documented, callers cannot predict return structure. Error handling is minimal; tools return plain strings without recovery guidance. No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels. Composition is sound: each tool has one responsibility, and tool names are unambiguous.
Execute a Neovim command and return the output.
Get the contents of a buffer.
Get diagnostics for a buffer.
Get recent messages from Neovim.
Get the current state of Neovim (buffers, windows, cursor position, etc.).
Run :checkhealth and return the output.
Check if Neovim is currently running.
Output schemas not documented. Tools return strings or unspecified structures; LLMs cannot predict fields for downstream chaining. E.g., nvim_get_state, nvim_get_buffer, nvim_health_check lack return type documentation.
Tool annotations missing. No readOnlyHint or destructiveHint despite clear risk levels (nvim_execute, nvim_lua, nvim_send_keys are WRITE; nvim_screenshot, nvim_get_* are READ_ONLY). Agents cannot infer safety/idempotency.
Parameter constraints under-specified. nvim_execute 'cmd' has no format guidance (e.g., ':set number' vs 'set number'). nvim_send_keys 'mode' lists options ('m', 'n', 't', 'x') but no explanation of when to use each. nvim_resize rows/cols lack min/max bounds.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2026-07-28+ | v2 |
Execute Lua code in Neovim and return the result.
Resize the Neovim terminal.
Take a screenshot of the Neovim window and save it as PNG.
Send keystrokes to Neovim.
Start an embedded Neovim instance. Always call first. Check nvim_is_running before starting a second instance.
Stop the running Neovim instance.
Inconsistent description depth. nvim_stop (5 words), nvim_is_running (8 words), nvim_health_check (9 words) are below 20-char minimum for actionable guidance. nvim_start (200+ chars) is well-detailed. Inconsistency confuses LLM tool selection.
Error handling lacks recovery guidance. Tools return plain error strings (e.g., 'Failed to start Neovim: {e}') without actionable next steps. No categorization of retryable vs fatal errors. LLMs cannot self-correct.