Docs
Agent capabilities

Agent capabilities

Combine connected software, durable memory, and CoreSpeed-hosted tools behind one caller-specific MCP surface.

CoreSpeed keeps capability setup outside the agent runtime. A client authenticates once, asks MCP for its visible tools, and receives only the tools available to that member and organization.

The v0 capability model has two layers:

Connected software                        CoreSpeed-hosted capabilities
OAuth account + member/org context        Memory and built-in tools
                 \                        /
                  POST /mcp → tools/list
Primary product surface
Connectors

Attach private or shared software accounts without handing OAuth tokens to an agent.

Durable context
Memory

Keep stable facts and decisions available across clients and conversations.

Execution contract
MCP

Discover and call the exact namespaced tool surface visible to this caller.

Visibility
Capability controls

Organization settings form the ceiling; members can narrow built-ins for themselves.

Hosted by CoreSpeed
Built-in tools

Treat tools/list as authoritative while the authenticated engine inventory is being completed.

V1 · Coming soon
Approval gate

Compile plain-English intent into policy code that runs before outbound actions.

Two discovery questions

  • GET /connectors answers what software can this caller connect, and what account state already exists?
  • MCP tools/list answers what can this caller invoke right now?

The retired public /catalog is not part of the v0 discovery contract.

On this page

Two discovery questions