Does DecisionPen store my AWS keys?

Not on the edge push path. Credentials stay on the MCP edge process. DecisionPen receives content and provenance after propose/sync.

Does fetching an S3 object update the Confirmed map?

No. get_document is cite-only. propose_enrichment and sync land in Inbox for human approval.

How do I connect DecisionPen MCP to an S3 document repository?

What this connection is for

Enterprise discovery and institutional memory often live in an object store: High Level Designs, packs, and exports under a known prefix. The Context Engine can read that allowlisted prefix and push selected files into DecisionPen for human review.

Important: on this path, DecisionPen does not hold your AWS keys. Credentials stay on the edge — your laptop, VPC, or container running the MCP server. That matches how agents sync document repositories into the Architecture Context Plane without turning DecisionPen into your IAM broker.

GitHub runbook: SYNC.md · MCP README. First connect Claude or Cursor with this setup guide (CONNECT.md).

What agents can do with S3 via MCP

list_sources — browse folders and files under the allowlisted prefix.

search_docs — search object keys.

get_document — fetch one object to cite in research ( does not persist into DecisionPen by itself).

propose_enrichment — download on the edge and push into DecisionPen for Inbox review.

run_sync — sync new or changed objects (also available as a CLI/cron job).

Agents can still use submit_document for text they already have — including content gathered from other MCP sources such as SharePoint in the agent host. Native SharePoint admin connectors are not claimed here.

Setup outline

Create an Edge API token (Step 1 in the connect guide).
Set DP_API_BASE, DP_API_TOKEN, and DP_MCP_MODE=propose when you are ready to push.
Set DP_S3_BUCKET and DP_S3_PREFIX (prefix is required for propose/sync).
Give the edge process IAM rights to s3:ListBucket and s3:GetObject only on that bucket/prefix.
Dry-run sync, then enable MCP in Claude or Cursor, or schedule daily sync. Details and Docker examples live in SYNC.md.

How this supports the Context Plane

Document repositories feed enterprise discovery and institutional memory. Combined with incremental modelling, agents propose structure from what you already store; stewards curate; living enterprise architecture stays current as new packs land — without another quarterly archaeology project.

What comes next

Continue the docs path with Beyond traditional EA, or deepen agent recipes in the GitHub MCP documentation.