Skip to content

Prism — Semantic Code Intelligence

Prism is a hosted MCP server that gives AI coding assistants deep, always-current semantic understanding of any codebase. Instead of an AI agent exploring files blindly and burning context tokens, Prism answers in milliseconds with exactly the right code chunks, structural maps, and symbol definitions.

Live gateway: https://prism-gateway-xvsemyikqq-oa.a.run.app MCP endpoint: https://prism-gateway-xvsemyikqq-oa.a.run.app/mcp

Key Components

Component Purpose
MCP Gateway Starlette ASGI app with Streamable HTTP MCP transport, dual auth (developer tokens + JWT), per-tenant routing, rate limiting
Hybrid Search Engine 4-way retrieval (vector + BM25 + description + exact) with RRF fusion and Google Semantic Reranker
VertexAI Embedder Generates 2000-dimensional code embeddings via gemini-embedding-001
pgvector Storage Per-tenant schema isolation in Cloud SQL PostgreSQL 16 + pgvector 0.8.1
AST Chunker Tree-sitter-based code chunking with symbol extraction across 48 languages
Ingestion Pipeline GitHub webhook-triggered indexing: clone, chunk, embed, store, rebuild PageRank map
Console API REST endpoints for repo management, developer tokens, and indexing controls
Job Queue Direct (asyncio) or Cloud Run Jobs dispatch for indexing work

MCP Tools

Tool Description
ping Connection health check
search_code Semantic + exact code search with intent classification and enriched responses
get_module_map PageRank-ranked structural map of modules and functions
get_symbol_definition Go-to-definition: file, line, signature, docstring
find_references All usages of a symbol grouped by type (definition, imports, calls)
get_recent_changes Recently indexed file changes

Documentation Sections

  • Overview — What Prism does, who it serves, capabilities, and limits
  • Architecture — System design, component responsibilities, data model, design decisions
  • Operations — Deployment, GCP infrastructure, monitoring, runbooks
  • Repo Onboarding — Step-by-step guide for connecting a repo and configuring AI clients