MCP server for energy infrastructure — dispatch batteries, verify carbon, read grid meters across 8 protocols
energyatit has 19 well-registered tools with consistent naming (verb_noun pattern) and reasonable descriptions (70-180 chars, baseline 194). However, schema quality is mixed: input parameters are properly typed with Zod and descriptions present, but output schemas are completely undocumented, tool responses are JSON blobs with no guidance on structure, fields, or chaining IDs. Error handling is basic (generic try-catch returning error text) with no recovery hints or categorization. Parameters are constrained where needed (command enum implied but not formally declared in schema), but lack dependency documentation and format guidance. No parameter validation rules stated. Security check: API keys are properly injected via environment variables, not exposed as parameters. Composition is adequate (each tool has one responsibility), but response structure makes chaining difficult, agent cannot know if downstream tools will receive required IDs. Tool design matches energy domain well (sites, assets, grid connections, dispatch, carbon) but lacks the polish of production tools (e.g., list_settlements description does not explain what a settlement is or when to generate one).
Create a carbon attestation record in the hash chain
Create a demand response event
Send a dispatch command to an asset (battery, HVAC, EV charger, etc.)
Execute dispatch for a DR event
Get dispatch history for an asset
Generate a hash-chained settlement for a site
Get carbon attestation for a site
Output schemas completely undocumented. Tools return text(await apiGet(...)) with no schema guidance, making it impossible for agents to know what fields to expect or how to chain results. All 19 tools return JSON via text() helper with zero documentation of response structure.
Destructive operations (dispatch_command, generate_settlement, create_carbon_record, create_dr_event, dispatch_dr_event, settle_dr_event) lack confirmation or dry-run capability. Agent could accidentally dispatch battery commands or create fake carbon records without user approval.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 66 | 2026-07-28+ | v2 |
Generate a carbon certificate for a facility
Get details of a DR event
Get meter readings for a grid connection
Get details of a specific site
List assets, optionally filtered by site
List demand response events
List grid connections for a site
List settlements for a site
List all energy sites in your tenant
Settle a DR event
Verify the SHA-256 hash chain for a meter
Verify a settlement's hash chain integrity
Error handling is generic. All tools catch errors and return errorResult(e) with only the error message text. No categorization (retryable vs user-fixable), no invalid-value echoing, no recovery hints. Agent cannot distinguish between 'retry this' and 'ask the user'.
Parameter 'command' in dispatch_command has description listing enum values ('charge, discharge, reduce, curtail, shed_load, restore') but schema does not enforce enum constraint. Zod z.string() allows any value; LLM could pass invalid commands.
Parameters lack format constraints and validation rules. Example: 'period_start' and 'period_end' in generate_settlement described as 'ISO date' but z.string() with no .regex() or .datetime() validation; agent could pass invalid dates. 'grid_zone' in create_carbon_record marked optional but no guidance on valid zones.
Descriptions for discovery tools (list_sites, list_assets, list_settlements, list_dr_events) lack pagination hints. No documentation of result limits, offset/limit parameters, or total_count field. Agent does not know if results are capped at 20 or 500 items.
Descriptions for write operations do not state consequences. E.g., dispatch_command description does not warn that commands actually control hardware (batteries, HVAC, EV chargers). Agent may not grasp irreversibility.
No tool-chaining IDs documented. E.g., list_sites response structure unknown, does it return site_id, id, or siteId? get_site then expects site_id as parameter. Mismatches force agent to guess field names.
Descriptions are domain-specific but lack LLM guidance. E.g., 'Generate a hash-chained settlement for a site' does not explain what a settlement is, why an agent would generate one, or when to call it instead of verify_settlement. Agent must infer purpose from context.