Your agent has tools. Does it have the right evidence? Connecting Jylus through MCP

Jylus, founded by an unnamed developer, has launched an evidence and state layer that connects to compatible AI agents through the Model Context Protocol (MCP), preparing source-backed context before a model reasons. The read-only MCP endpoint exposes three tools, including get_context_pack, which returns supporting source records so developers can check whether evidence establishes a claim, contains unresolved disagreements, or is missing required records. The founder recommends testing by keeping a question identical while changing one underlying record to verify the returned evidence reflects the change.

Your agent successfully calls three tools. The CRM says Enterprise. Billing says the customer downgraded yesterday. The support document describes Enterprise benefits. Every call succeeded. The agent can still give the wrong answer. Connecting tools gives an agent access to information. Working out which facts apply together is another part of the system. I’m the founder of Jylus. We built an evidence and state layer that prepares source-backed context before a model reasons. It now connects to compatible agents through MCP. Here’s what that means for developers. What MCP changes Model Context Protocol gives AI applications a standard way to discover and call tools exposed by a server. For Jylus, that means a compatible client can request evidence through the MCP connection without you writing a separate tool wrapper around each Jylus API operation. Your model handles reasoning. Jylus prepares the evidence it receives. Three tools, with distinct jobs The connection is read-only. Your application or ingest pipeline supplies the underlying data separately. A practical first request After ingesting your records, ask your agent to call "get context pack" with arguments like these. Replace the namespace and dates with your own: { "question": "What is customer C-104's current plan, and which records support it?", "namespace": "customer-events", "from": "2026-10-01T00:00:00Z", "to": "2026-10-08T23:59:59Z", "token budget": 2000 } Then ask it to inspect the supporting source records. Does the evidence establish the plan? Is there an unresolved disagreement? Is a required record missing? Those are useful things to inspect before judging the final prose. Connecting it The remote endpoint is: " https://api.jylus.ai/api/mcp https://api.jylus.ai/api/mcp " It uses Streamable HTTP and a workspace API key with "events:read". Keep the key in private client settings or a secret store. Client-specific setup: https://jylus.ai/mcp https://jylus.ai/mcp Connecting the server does not automatically capture your chat history or upload your project. It gives the client tools to query data already retained in your Jylus workspace. The test I’d start with Keep the question identical. Change one underlying record. Ask again. Check whether the returned evidence reflects the change, whether the supporting source IDs make sense, and whether incomplete evidence remains visible to the model. You can explore the Context Packs first in the browser playground: https://jylus.ai/try https://jylus.ai/try For developers using MCP: what takes more work today—connecting another tool, or making the information from your existing tools agree?