Summary
Add two new block types to Voiden that connect its existing request-building workflow to AI agents. This is two separate pieces of work; /mcp ships first, /tool after.
Block 1: /mcp
A new request type (alongside REST/GraphQL/WebSocket) for connecting to and testing someone else's MCP (AI-agent) server — via stdio or http — so agent servers can be verified, version-controlled, and run in CI like any other request. Captures connection details (transport, command/args or URL, env, auth), an operation (list / call tool / read resource / get prompt), and assertions on the result.
Block 2: /tool
A marker attached to an existing Voiden request that exposes it as a named, described capability for an AI agent to call directly, instead of today's four generic actions. Captures identity (name, title, description, annotations), parameters (with a source of agent vs environment to keep secrets safe), and a verification policy (linked test requests, cadence, mode, on-failure behavior) that determines whether the tool shows as verified, unverified, or failing.
Rollout order
/mcp — simpler, reuses existing infrastructure./tool — requires new infrastructure to let the agent connection read a /tool block as a live capability.
Summary
Add two new block types to Voiden that connect its existing request-building workflow to AI agents. This is two separate pieces of work;
/mcpships first,/toolafter.Block 1:
/mcpA new request type (alongside REST/GraphQL/WebSocket) for connecting to and testing someone else's MCP (AI-agent) server — via stdio or http — so agent servers can be verified, version-controlled, and run in CI like any other request. Captures connection details (transport, command/args or URL, env, auth), an operation (list / call tool / read resource / get prompt), and assertions on the result.
Block 2:
/toolA marker attached to an existing Voiden request that exposes it as a named, described capability for an AI agent to call directly, instead of today's four generic actions. Captures identity (name, title, description, annotations), parameters (with a source of
agentvsenvironmentto keep secrets safe), and a verification policy (linked test requests, cadence, mode, on-failure behavior) that determines whether the tool shows as verified, unverified, or failing.Rollout order
/mcp— simpler, reuses existing infrastructure./tool— requires new infrastructure to let the agent connection read a/toolblock as a live capability.