Loading ToolBench
MCP server for Agentforce Grid — exposes the Grid Connect API to LLM agents (Claude Code, Cursor, ChatGPT desktop) via the Model Context Protocol
Critical gaps in schema visibility and parameter documentation. Most tools use generic action-based dispatch (workbook, worksheet, column, cell, discover) with a single 'action' string parameter, making schemas opaque to LLMs. Tool descriptions are present but lack specificity about when to use each action variant. No output schemas documented. Error handling guidance absent. Only apply_grid and poll_worksheet_status have reasonably complete parameter definitions.
Create or update an Agentforce Grid from a YAML specification. Accepts a declarative YAML string describing the grid structure (workbook, worksheet, columns, data). Creates all resources in dependency order, resolving column references automatically.
Cell operations: update, paste, trigger_execution, validate_formula, generate_ia_input
Column CRUD operations: add, edit, save, delete, reprocess, get_data, create_from_utterance, generate_json_path
Column mutation typed shorthands: edit_ai_prompt, edit_agent_config, add_evaluation, change_model, update_filters, reprocess_typed, edit_prompt_template
Metadata/data/agent discovery actions (23+ discovery operations)
Get URL for a grid resource
Seven tools (workbook, worksheet, column, column_mutation, cell, discover, get_worksheet_summary) use generic 'action' string dispatch with no enum constraints or per-action schema variants. LLMs cannot determine valid action values or required parameters for each action without reading source code.
No output schemas documented for any tool. LLMs cannot plan downstream tool calls or extract required fields (e.g., worksheetId, columnId) from responses without trial-and-error.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | 2026-07-28+ | v2 |
Get worksheet summary information
Poll worksheet status until cells reach terminal state
Set up a complete agent test workbook with utterances and evaluations
Workbook operations: list, create, create_with_worksheet, get, get_worksheets, delete
Worksheet operations: create, get, get_data, get_data_generic, update, delete, add_rows, delete_rows, import_csv, run, get_run_job
Tool descriptions lack specificity about when to use each action variant. E.g., 'Worksheet operations: create, get, get_data, ...' lists actions but does not explain when to call get_data vs get_data_generic or when to use import_csv vs add_rows.
No error handling guidance. Tools do not document what errors are retryable, what user-fixable errors look like, or what recovery steps to take. Agents will not know whether to retry, ask the user, or fail.
apply_grid accepts a 'spec' parameter as a raw YAML string with no schema validation hints. LLMs cannot know what YAML structure is valid without reading documentation or trial-and-error. No dryRun output schema documented.