A Model Context Protocol server for a CPGRAMS (Centralized Public Grievance Redress and Monitoring System) simulation portal. Provides tools for citizens to file, track, rate, and appeal grievances through an AI agent interface with human approval gates.
Strong tool definitions with comprehensive descriptions (avg 250+ chars), well-structured schemas, and clear parameter types. All 13 tools have explicit input schemas with proper JSON Schema format. Tool names follow verb_noun convention (get_, list_, create_, update_, send_). Error handling includes actionable recovery guidance. Main gaps: output schemas are documented only in descriptions (not as formal outputSchema), some parameter descriptions could be more prescriptive about constraints, and write tools lack explicit confirmation/dry-run patterns despite being marked as requiring approval.
Start an appeal draft for a disposed grievance. Input: {grievanceId, grounds, argument}. Checks eligibility (status DISPOSED, not yet appealed, rating is Poor). Returns JSON {ok, speakable, data:{appealDraft:{grievanceId, grounds, argument}}, nextActions}. Write; requires approval.
Start a new grievance draft. Input: {categoryId, subject, description, reliefRequested, evidence?:[{name, mimeType, base64}]}. Validates category and draft shape. Returns JSON {ok, speakable, data:{draft:{id, categoryId, subject, ...}}, nextActions}. Write; requires approval.
Snapshot of the citizen's case workspace: current screen, language, case counts by status, which cases need attention today (SLA overdue / unrated / appeal-eligible), and which write actions are currently available. Start here. Returns JSON {ok, speakable, data:{view, lang, counts, attention:[{regId,subject,why}], can:{...}}, nextActions}. Read-only.
Full record of one grievance by registration ID (e.g. PG-26-03877): subject, description, relief requested, ministry, status, evidence list, interim reply, reminders, disposal, rating, appeal, and the movement timeline. Input: {grievanceId}. Returns JSON {ok, speakable, data:{grievance}, nextActions}. Read-only.
Output schemas documented only in descriptions, not as formal outputSchema fields. LLMs cannot parse unstructured text descriptions to infer response structure reliably.
Write tools (create_grievance_draft, update_grievance_draft, submit_grievance, send_reminder, rate_disposal, create_appeal_draft, send_appeal) marked as requiring approval but no explicit confirmation/dry-run mechanism visible in tool definitions. Approval infrastructure exists in code (checkGate, consumeApproval) but not surfaced as tool parameters or Multi-Round-Trip Request pattern.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | 2025-06-18+ | v2 |
Knowledge base: rules, timelines, and FAQs about the CPGRAMS process. Input: {query} (e.g. 'How long does redressal take?', 'Can I appeal?', 'What is an interim reply?'). Returns JSON {ok, speakable, data:{answer, sources:[{title, excerpt}]}, nextActions}. Read-only.
SLA tracking for all grievances: days elapsed, redressal target, status (on-track / overdue), attention reason (if any), and eligibility for reminder / rating / appeal. Returns JSON {ok, speakable, data:{cases:[{regId, subject, status, daysElapsed, target, onTrack, attentionReason, canRemind, canRate, canAppeal}]}, nextActions}. Read-only.
Categories this simulation accepts, routed to Central Ministries/Departments (Railways, EPFO, Health, Education, Power, Consumer Affairs) with a 21-day redressal target. Titles return in ONE locale (default en; pass lang:'hi' for Hindi). Use before filing so category ids are valid. Returns JSON {ok, speakable, data:{categories:[{id,title,ministry,needsEvidence}],redressalTargetDays}, nextActions}. Read-only.
Rate a disposed grievance (Satisfactory / Average / Poor). Input: {grievanceId, rating}. Checks eligibility (status DISPOSED, not yet rated). Returns JSON {ok, speakable, data:{grievance:{regId, rating, status}}, nextActions}. Write; requires approval.
File the active appeal draft with the Nodal Appellate Authority. Input: {} (no parameters). Validates appeal completeness and transitions the grievance to APPEAL_PENDING status. Returns JSON {ok, speakable, data:{grievance:{regId, appeal:{status, filedAt, grounds, argument}}}, nextActions}. Write; requires approval.
Send a reminder to the ministry for a grievance in UNDER_PROCESS status. Input: {grievanceId}. Checks SLA eligibility (≥7 days since filing or last reminder). Returns JSON {ok, speakable, data:{grievance:{regId, reminders:[...]}}, nextActions}. Write; requires approval.
Optional voice narration (accessibility). Input: {text, lang?:'en'|'hi'}. Speaks the text aloud in the chosen language if Voice Mode is on; otherwise silent. Returns JSON {ok, speakable, data:{spoken:true}, nextActions}. Read-only.
File the active draft as a new grievance. Input: {} (no parameters). Validates draft completeness, assigns registration ID (PG-26-XXXXX), records filing timestamp, and transitions to RECEIVED status. Returns JSON {ok, speakable, data:{grievance:{regId, subject, status, filedAt, ...}}, nextActions}. Write; requires approval.
Edit an active grievance draft. Input: {subject?, description?, reliefRequested?, evidence?}. Partial update; omitted fields are unchanged. Returns JSON {ok, speakable, data:{draft}, nextActions}. Write; requires approval.
Parameter descriptions for evidence arrays (create_grievance_draft, update_grievance_draft) lack constraints on file size, count, or supported MIME types. LLMs may attempt to upload oversized or unsupported files.
update_grievance_draft has no required parameters, all fields are optional. Description states 'Partial update; omitted fields are unchanged' but does not clarify whether calling with empty object {} is valid or if at least one field must be provided.
Authorization gate (requireCitizen) enforced in tool implementations but not documented in tool descriptions. LLMs cannot predict which tools require sign-in without calling them first.