Investigate supporting Pinecone MCP Server

[tractorjuice]

https://docs.pinecone.io/guides/operations/mcp-server

[tractorjuice]

Investigation Findings

What Pinecone MCP Provides

Pinecone is a vector database service. Its MCP server (installed via npx @pinecone-database/mcp) offers 9 tools:

Tool	Purpose
search-docs	Search Pinecone documentation
list-indexes	List available indexes
describe-index	Get index details
describe-index-stats	Get index statistics
create-index-for-model	Create new indexes
upsert-records	Insert/update vector records
search-records	Query with metadata filtering & reranking
cascading-search	Search across multiple indexes
rerank-documents	Rerank results using specialized models

Potential Fit with ArcKit

• Could enhance research agents (datascout, research) with semantic search over architecture artifacts
• Could enable searching across project documents using vector embeddings
• Vector search could improve traceability chain lookups

Challenges

• Requires a Pinecone API key (PINECONE_API_KEY) — unlike AWS Knowledge and Microsoft Learn MCPs which are free/public
• Requires Node.js (npx) to run the server
• Only supports indexes with integrated embedding — this feature is still in early access, not production-ready
• The existing bundled MCPs (AWS, Microsoft, Google) are documentation-search tools — Pinecone is fundamentally different (a data platform)
• No obvious mapping to a specific ArcKit command or agent pattern

Recommendation

Worth monitoring but not ready for integration yet because:

1. Cost barrier — requires a paid Pinecone account, unlike the free documentation MCPs
2. Early access — the integrated embedding requirement is still in early access
3. Unclear use case — no specific ArcKit command naturally consumes Pinecone tools
4. User configuration — would require users to provide their own API key, unlike the zero-config bundled MCPs

A more natural integration point might emerge if ArcKit adds features for searching/querying across generated architecture artifacts, or if a future command needs semantic similarity search.

Suggest keeping this issue open for periodic review.

[tractorjuice]

Additional Thought: Document Ingestion & Semantic Search

One compelling use case would be a capability to ingest documents into Pinecone, enabling semantic search across an organisation's entire knowledge base — not just ArcKit artifacts, but any existing documentation.

Concept: /arckit.index command

A new command that:

1. Scans configurable source directories for documents (ArcKit projects/, but also existing docs, policies, strategies, runbooks, wiki exports, etc.)
2. Chunks each document intelligently (respecting section boundaries, headings, metadata)
3. Upserts them into a Pinecone index with rich metadata (source, document type, classification, status, version)
4. Enables subsequent semantic queries across the full document corpus

What This Would Unlock

Across ArcKit artifacts:

• Cross-project traceability — "Which projects have requirements related to data sovereignty?" across dozens of projects
• Impact analysis — "What stakeholders and risks are affected by a change to the authentication service?"
• Duplicate/conflict detection — Find overlapping or contradictory requirements across projects
• Architecture portfolio search — "Show me all ADRs related to cloud migration decisions"
• Governance reporting — Semantically query across all artifacts for audit and compliance reviews

Across existing organisational documents:

• Policy alignment — Check new architecture decisions against existing corporate policies, security standards, and governance frameworks
• Knowledge discovery — Surface relevant existing documentation (design docs, post-mortems, runbooks) when starting a new project
• Gap analysis — Identify areas where architecture documentation is missing by comparing against existing operational docs
• Onboarding — New architects can semantically search across the entire document corpus to understand existing decisions and context
• Regulatory compliance — Cross-reference architecture artifacts against existing compliance documentation, audit reports, and regulatory guidance

How It Could Work

/arckit.index                              # Index all ArcKit artifacts in current project
/arckit.index --all                        # Index all ArcKit projects
/arckit.index --source ./docs              # Index an arbitrary docs directory
/arckit.index --source ./policies --tag compliance   # Index with custom tags
/arckit.index --query "GDPR"              # Search across indexed documents

Each document would be stored with metadata filters:

• source: e.g. "arckit", "policies", "runbooks", "wiki"
• project_id: e.g. "001", "002" (for ArcKit artifacts)
• doc_type: e.g. "REQ", "ADR", "RISK", "STKE", "policy", "runbook"
• classification: e.g. "OFFICIAL"
• status: e.g. "APPROVED", "DRAFT"

This would allow filtered searches like "find all APPROVED requirements and related policies that mention API integration".

Feasibility Concerns

• Still dependent on users having a Pinecone account and API key
• Document chunking strategy would need careful design — ArcKit's structured templates are easier to chunk than arbitrary docs
• Would need to handle re-indexing when documents are updated (version tracking)
• Arbitrary document ingestion increases scope significantly but also increases value
• Could be a significant value-add for organisations with large architecture portfolios (e.g. the UK Gov department repos v23–v43) and existing document libraries

This use case strengthens the argument for eventual Pinecone integration, but the prerequisite issues (cost barrier, early access status) still apply.