Loading ToolBench
MCP server for synthesizing software engineering patches using Agent Computer Interface (ACI) navigation and automated patch generation
This MCP server has a single tool, execute_skill_action, which is extremely underdeveloped. The server lacks any formal tool registration with JSON Schema, has no input parameter documentation, no output schema definition, and the tool description is generic and too brief (50 chars). The implementation is essentially a stub that prints a JSON message claiming it's 'ACTIVE_LISTENING' but provides no actual MCP protocol compliance. There is no evidence of proper parameter validation, error handling, or structured tool definitions. The source code shows only a mock client and example usage, no real MCP message handling, no tool invocation logic, and no response formatting. This is not a functional MCP server; it's a skeleton.
Execute a skill action for SWE patch synthesis
No input schema defined for execute_skill_action. Tool registration lacks JSON Schema with parameter types, constraints, and validation rules. Critical for LLM tool selection and parameter binding.
Tool description is only 50 characters ('Execute a skill action for SWE patch synthesis'), too brief to guide LLM selection. Fails baseline requirement of 34 - 392 characters. Missing: what the tool does, when to use it, what it returns, prerequisites.
No parameter documentation. The tool accepts some form of input (inferred from the client example: issue_description, target_repo_path) but none of these parameters are formally defined with types, descriptions, constraints, or enums.
No output schema documented. The client.py mock shows a hardcoded response structure (patch_session_id, target_repo, aci_navigation_commands_executed, diff_patch_synthesized_lines, regression_test_suite_passed, swe_benchmark_resolved, patch_diff_url) but this is not formally declared in the MCP tool definition.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 16 | 2026-07-28+ | v2 |
Tool name 'execute_skill_action' is vague and generic. Does not follow verb_noun pattern clearly ('execute' is too broad). 'synthesize_swe_patch' or 'generate_patch_for_issue' would be more specific and aid LLM disambiguation.
No error handling guidance. Tool description does not indicate what errors are retryable, what are user-fixable, or what recovery actions (e.g., search for similar issues) the LLM should take.
Destructive operation (WRITE risk) without confirmation pattern. The tool synthesizes and applies patches, an irreversible operation. No dry-run, preview, or confirmation mechanism documented.
Server does not implement actual MCP protocol. The mcp_server.py prints a static JSON message and exits. No message loop, no tool invocation handling, no proper request/response framing per MCP spec.