MCP server for controlling Alacritty terminal instances, capturing screenshots, sending keys, and extracting Neovim context
This server defines 5 tools with basic JSON-RPC handling and STDIO transport. Tool definitions are visible in src/mcp_server.rs with names, descriptions, and input schemas present. However, significant quality gaps exist: (1) Descriptions are generic and lack actionable guidance on WHEN and WHY to use each tool; (2) Input schemas have type definitions but lack detailed parameter-level descriptions in most cases; (3) No output schemas are documented, responses are formatted as simple text strings rather than structured objects; (4) Error handling returns generic error codes without recovery guidance; (5) No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk classifications; (6) Parameters lack constraints and detailed format specifications; (7) Multi-step composition is not guided (e.g., spawn_instance then send_keys requires manual ID tracking). The server is functional but would not pass code review for production use.
Extract context from a Neovim instance running in an Alacritty terminal
List all running Alacritty terminal instances
Take a screenshot of an Alacritty terminal instance
Send keyboard input to a running Alacritty instance
Spawn a new Alacritty terminal instance
Tool descriptions lack WHEN/WHY guidance and actionable context. E.g., 'Send keyboard input to a running Alacritty instance' does not explain when to use send_keys vs other input methods, what xdotool format means, or what happens on failure.
Parameter descriptions are missing or minimal. 'keys' parameter in send_keys says 'Keys to send (xdotool format)' but does not explain what xdotool format is, valid examples, or constraints. 'instance_id' in multiple tools lacks format/encoding details.
No output schemas documented. All tools return responses formatted as simple text strings (e.g., 'Found X Alacritty instances:\n{json}') rather than structured JSON objects with typed fields. LLMs cannot reliably parse or chain these outputs.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 52 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 42 | 2024-11-05+ | v1 |
Error responses provide only generic error codes (-32602, -32603) without recovery guidance. E.g., if send_keys fails because instance_id is invalid, the response should suggest 'Call list_instances() to find valid instance IDs' rather than bare error code.
No tool annotations present despite clear risk classifications (WRITE, READ_ONLY). Tools should declare destructiveHint=true (spawn_instance, send_keys) and readOnlyHint=true (list_instances, screenshot_instance, get_neovim_context) to signal to agents which operations are idempotent and which have side effects.
Parameters lack type/format constraints. 'context_lines' in get_neovim_context is an integer but no min/max bounds specified. 'format' in screenshot_instance has enum ['text', 'image'] but no description explaining what each produces or when to use each.
No guidance on composition or multi-step workflows. spawn_instance returns an instance, but how does the LLM know to extract instance_id for use in send_keys or screenshot_instance? Output format is ambiguous text, not structured JSON with explicit instance_id field.
Destructive operations (spawn_instance, send_keys) lack confirmation or dry-run patterns. An agent could accidentally spawn dozens of terminal instances or send unintended keystrokes without a safety net.