Loading ToolBench
Local MCP server for FiveM development. Runs on the user's machine and connects to their AI client over MCP. All tools — file I/O, SSH, MySQL, txAdmin/custom panel, and the analysis suite (validation, security, event/dependency tracing) — run locally.
FiveClaw Agent provides 36 tools with basic descriptions and input schemas, but exhibits significant quality gaps. All tools have descriptions (positive), but many are generic or lack actionable detail. Input schemas are present but inconsistently documented, most parameters lack type constraints (enums, ranges, patterns). Error handling guidance is absent; tools do not explain recovery paths or categorize failures. Parameter descriptions are often minimal (e.g., 'Resource name' without format/constraint details). No output schemas are documented, forcing LLMs to infer response structure. Tool naming is generally clear (verb_noun pattern), but composition issues exist: repomap_generate, repomap_query, repomap_show are tightly coupled discovery tools that could be unified. Security concerns: tool_mysql_query and tool_ssh_exec expose raw command execution without validation hints or permission gates. No tool annotations (readOnlyHint, destructiveHint) despite 14 WRITE-risk tools. Baseline: 549 production tools average 194-char descriptions; FiveClaw averages ~100 chars, often too terse.
Analyze the structure of data passed between resources (tables, JSON, etc.).
Analyze how exports from a resource are used across the codebase.
Create a backup of a resource.
Load saved context/notes.
Save context/notes about the current project for later retrieval.
Deploy a resource to the FiveM server via SSH or local copy.
Scan a resource for common anti-patterns and code smells.
No output schemas documented. LLMs cannot infer response structure, forcing them to guess field names and types. This breaks tool chaining and increases hallucination risk.
14 WRITE-risk tools (tool_mysql_query, tool_ssh_exec, tool_server_console, deploy_resource, etc.) lack destructiveHint annotations and error recovery guidance. No confirmation step or dry-run option for irreversible operations.
Parameter descriptions lack actionable constraints. 'Resource name' does not specify format, length, or valid characters. 'SQL query' does not warn about injection risks or validation. LLMs cannot self-correct invalid input.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | <=2025-11-25 | v2 |
Find duplicate code blocks within a resource.
Find all event handlers registered in a resource.
Find all exports defined in a resource.
Find all event triggers (TriggerEvent, TriggerServerEvent, etc.) in a resource.
Search FiveM documentation (requires FiveClaw API key).
Look up FiveM native function documentation (requires FiveClaw API key).
Fetch client-side logs from txAdmin live console or custom panel.
Check FiveClaw Agent health and verify your connection is working. Returns environment status, tool count, and API key validity.
List all configured MySQL databases (primary + extras).
Generate boilerplate code from a pattern template (fivem-resource, mysql-storage, state-bag, qb-resource, etc.).
Read the latest FiveM server logs (server.log, console.log, or custom log file).
Scan your FiveM resources directory and build a map of all resources, their files, exports, and event handlers. Run this first.
Query the cached resource map by type (exports, events, files, or all) with optional resource name filter.
Display the full cached resource map.
Scan a resource for security vulnerabilities (SQL injection, XSS, unsafe patterns, etc.).
Get file metadata: size, line count, last modified time.
Execute a raw SQL query against the configured MySQL database.
Return the configured OS and environment paths for this machine.
Start, stop, or restart a resource on the FiveM server.
Search for a regex pattern in resource files (Lua, JS, TS, JSON).
Send a command to the FiveM server console and get the response.
Start or stop the entire FiveM server.
Get the current FiveM server status (running, stopped, etc.) from txAdmin or custom panel.
Download a file from the remote FiveM server via SSH.
Execute a shell command on the remote FiveM server via SSH.
Upload a file to the remote FiveM server via SSH.
Check Lua file syntax (CfxLua-aware). Uses luac binary if available, otherwise lupa (embedded LuaJIT) or luaparser.
Trace the flow of an event through the codebase (where it's triggered, handled, etc.).
Validate that exports are called with the correct number and types of arguments.
Three tightly coupled repomap tools (generate, query, show) should be unified or clearly documented with sequencing hints. LLMs may call them in wrong order or redundantly.
No error handling guidance. Tools do not explain what to do on failure (retry, ask user, abort). Raw error messages provide no recovery path for LLMs.