Loading ToolBench
Gephi AI provides 46 tools with basic descriptions and schema structures, but suffers from significant gaps in definition quality. While tool names follow the verb_noun convention adequately (get_, create_, delete_, run_, etc.), descriptions are inconsistent in depth and LLM-optimization. Input schemas are present for most tools but lack comprehensive parameter descriptions and validation constraints. Many parameters have minimal or absent descriptions. Error handling guidance is not visible in the schema structures provided. The server appears to expose a complex graph manipulation API but does not optimize parameter naming for natural-language agent interaction, e.g., requiring node_id or workspace_name instead of accepting human-friendly identifiers. Output schemas are not documented in the source provided. The definition quality reflects a functional but not production-optimized tool suite.
Apply a filter to the graph.
Remove multiple nodes from the graph.
Remove all nodes and edges from the graph.
Get value frequency distribution for a column.
Compare structural properties of two nodes.
Create a new project in Gephi.
Delete a workspace from the current project.
Parameter descriptions are minimal or missing. Many parameters like 'query', 'layout_name', 'statistic_name', 'filter_name', 'operation' lack substantive descriptions explaining their expected format, valid values, or constraints. This forces LLMs to guess acceptable input formats.
No output schemas documented. For all 46 tools, the source code does not show what fields are returned or how results are structured. This prevents LLMs from planning downstream tool calls or knowing what data to extract from responses.
Parameters require opaque system identifiers (node_id, workspace_name) without accepting human-friendly identifiers. Tools should accept both IDs and natural names, resolving them server-side. E.g., 'Remove node called Alice' should work alongside 'Remove node U2405687'.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 59 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 23 | - | v1 |
Detect potential duplicate nodes based on attributes.
Extract the backbone of the network using disparity filter.
Extract the ego network of a node (the node and its neighbors).
Extract the largest connected component from the graph.
Filter the graph to show only nodes within a degree range.
Filter the graph to show only edges within a weight range.
List all available layout algorithms.
List all columns (attributes) available for nodes or edges.
Get comprehensive statistics about the graph.
Determine if the graph is directed or undirected.
Get the configuration properties of a layout algorithm.
Get the status of the current layout algorithm.
Get details about a specific node.
Get the current visualization perspective/camera position.
Get the preview settings for graph rendering.
Get metadata about the current Gephi project.
Get the currently selected nodes and edges.
Get the timeline data if the graph is temporal.
Health check of Gephi Desktop and this server. Returns version, listening address, and a freshness check against the latest published versions.
List all available filters that can be applied to the graph.
List all available statistics that can be computed on the graph.
List all workspaces in the current project.
Merge multiple nodes into a single node.
Open an existing project in Gephi.
Profile the graph structure for patterns and anomalies.
Query edges by attribute or pattern.
Query nodes by attribute or pattern.
Remove an edge from the graph.
Remove isolated nodes (nodes with no edges) from the graph.
Remove a node from the graph.
Reset all node and edge colors, sizes, and other appearance properties.
Reset all filters to show the full graph.
Run a layout algorithm on the graph.
Run a statistic computation on the graph.
Take a manual snapshot of the current workspace for undo.
Stop the currently running layout algorithm.
Undo the last destructive operation by restoring from the auto-snapshot.
Generate and view an image of the current graph.
Perform what-if analysis on graph modifications without applying them.
No validation constraints visible for string parameters. Query expressions, filter names, and layout names lack format specifications, pattern rules, or enums. This invites invalid input from LLMs.
Destructive operations lack confirmation or dry-run support. Tools like gephi_clear_graph, gephi_remove_node, and gephi_delete_workspace can permanently destroy data without agent confirmation. No indication of undo/snapshot workflow in descriptions.
No error handling guidance visible. Descriptions do not indicate what can go wrong, when to retry, or what the user should do if a tool fails. E.g., gephi_open_project does not document what happens if the file is not found.
Tool names use domain-specific terminology (gephi_extract_backbone, gephi_profile_graph) that is not self-documenting. LLMs may not understand what 'backbone', 'profile', or 'whatif' mean without external Gephi knowledge.