MAPLE MCP server creation for RAG workflows

[nesanders]

MAPLE MCP Server & RAG Infrastructure

This PR implements a comprehensive Model Context Protocol (MCP) server for the MAPLE platform and upgrades the platform's vector search infrastructure to support Gemini-based RAG workflows.

Key Components

1. Model Context Protocol (MCP) Server

• New Package: Introduced a standalone MCP server in the mcp-server/ directory.
• Unified Search Tools: Implemented semantic search tools for:
  • Bills: Search legislative bills with filters.
  • Testimony: Search published testimony with billId or ballotQuestionId filtering.
  • Ballot Questions: Search ballot questions.
  • Policies: Unified search across multiple policy types.
• Authentication: Secured the server with a Firestore-based API key layer (agentKeys collection).
• Deployment Ready: Configured for Google Cloud Run with .env.example and optimized dependencies.

2. Gemini Embedding 2 Migration

• High-Dimension Vectors: Migrated all vector indexing logic from 768 to 3072 dimensions using gemini-embedding-2.
• Dynamic Response Handling: Updated all embedding utilities to robustly handle Vertex AI's evolving API response structures (embedding vs embeddings).
• Instruction-Tuned Prefixing: Implemented the required instruction-tuned prefixes for queries (task: search result | query: ...) and documents (title: ... | text: ...).

3. Backend & Data Infrastructure

• Real-time Indexing: Updated the Cloud Functions for bills, testimony, and ballot questions to use the new embedding model and dimensions.
• Composite Vector Indexes: Added the necessary Firestore index configurations in firestore.indexes.json to support filtered vector queries.
• Administrative Tools:
  • Backfill Script: Created backfill-embeddings.ts for batch migrating existing documents to the 3072-dim format.
  • Sync Utility: Updated syncBallotQuestions.ts for data consistency.

4. Developer Experience & CI/CD

• Environment Synchronization: Updated package.json and yarn.lock (root and functions) to resolve dependency conflicts.
• Type Safety: Resolved project-wide TypeScript deprecations and Node.js type mismatches.
• Coding Standards: Applied Prettier formatting and ESLint fixes across all new and modified files.

Status

✅ Completed

• MCP Server core implementation and tool registration.
• Transition to Gemini Embedding 2 (3072 dimensions).
• Updated Cloud Functions for real-time vector indexing.
• Firestore Index configurations for DEV.
• Resolution of all CI/CD blockers.

⏳ Pending Actions (Post-Merge)

• Index Deployment: Run firebase deploy --only firestore:indexes in the target environment.
• Data Migration: Execute the backfill script to populate 3072-dim vectors for existing records.
• Container Deployment: Deploy the MCP server image to Cloud Run.

---

This PR represents the foundational infrastructure for AI-driven legislative analysis on the MAPLE platform.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
maple-dev	Ready	Preview, Comment	May 15, 2026 5:27pm