Give your agent
the map.
Claude, Cursor, Cline, your in-house coding agent. Connect any of them to the live blastmap model over MCP. 12 read-only tools, every answer grounded in real edges with confidence and provenance.
Two ways. Pick one.
Add blastmap from the Claude Code marketplace. Auth handled, tool list registered, ready to query.
claude marketplace add blastmap-dev/blastmap-agentsPoint Claude at your already-published graph over MCP. No daemon.
claude mcp add blastmap \
--stdio \
--command "bunx blastmap mcp"Talk to the system in plain English.
Ask the question, the agent calls the right tool, the model returns evidence with confidence, the agent composes a grounded reply. No invented edges, no false certainty.
12 read-only tools over the model.
Read-only on purpose. The agent describes, queries, walks, ranks. It does not mutate state. Writing is a separate, narrower surface earned at higher trust.
Other clients. Same model.
MCP is a contract, not a vendor. Wire any client that speaks it.
Point any MCP-capable client at the published graph.
claude mcp add blastmap \
--sse \
--url https://graph.acme.dev/mcp \
--auth $BLASTMAP_TOKENVerify the graph data against live mcp-client data.
cursor mcp add blastmap \
--stdio bunx blastmap mcpAn agent that is confidently wrong is worse than none.
The Agent API exposes the discipline the model is built on. Your agent inherits the receipts.
Every tool result carries the extractor and the location. The agent can cite the edge back to a line.
Every edge carries a number in [0, 1]. The agent must rank and gate on it instead of pretending all knowledge is equal.
Every fact knows when it was last verified. Stale answers do not get to pretend they are current.