What the Context Engine is
Once you have an Architecture Decision Workspace, the next question is how agents join it. DecisionPen ships an MCP server we call the Context Engine.
Claude, Cursor, and your SDLC agents can use it to treat the Architecture Context Plane as shared truth. Delivery and design stay on the same plane as enterprise architecture. Nobody has to paste the estate into every prompt.
Technical tool list and recipes also live on GitHub: decisionpen_mcp README.
What agents can do
Through MCP, an agent can search entities and read relationships on the curated map. It can pull evidence and derived architecture-decision cards. It can inspect open estate changes. It can ask questions against structured context — not against file similarity alone. For interrogation inside the product, see Enterprise Ask.
With an allowlisted object store, agents can list, search, and fetch documents, then propose enrichment. Cursor or Claude can also orchestrate other MCP sources and push text with submit_document. See Connect S3 documents.
A typical publish loop: the agent sends text or an S3 object, evidence is indexed, proposals enter Inbox, and the graph updates only when a human approves. That is how agentic SDLC stays two-way.
How to connect
Create an Edge API token under Admin → Security & audit → Edge API tokens, install the MCP server in a dedicated venv, then wire Cursor or Claude Desktop. Full steps: Connect DecisionPen MCP to Cursor or Claude.
What comes next
Next: follow the setup guide — Connect Cursor or Claude.