ConveRact · guides
Connect ConveRact agents with MCP and A2A
ConveRact connects approved external tools and agents through Model Context Protocol (MCP) and Agent2Agent (A2A). Administrators manage connections, review available tools, grant access to selected agents, and inspect invocation activity from the Interoperability workspace.
Discuss your agent integrationFree integration. No installation charge or one-time ConveRact fee. Plan, usage, third-party, and custom development costs may apply.
Reviewed · ConveRact team
Use ConveRact as an MCP client
Register an external MCP server with its HTTPS endpoint and credentials. Discover its tools, review their descriptions and input schemas, and enable only the tools your workspace needs. Discovery does not automatically authorize every tool: changed schemas require another review. Calls use the configured external server and remain subject to its authentication and availability.
Expose a ConveRact agent as an MCP tool
Create a publication for an approved ConveRact agent and issue a scoped, expiring access grant. External MCP clients can discover the permitted ask_agent tool and request answers grounded in that agent?s configured knowledge. This interface provides agent answers; it does not grant unrestricted access to customer records or authorize transactions. The implementation supports MCP version 2025-11-25 over a stateless HTTP endpoint.
Connect agents through A2A
Register a remote A2A agent to call it from the workspace, or publish a ConveRact agent for an authorized external caller. The current implementation supports A2A version 0.3.0 with synchronous message/send requests. Streaming, asynchronous task subscriptions, and an autonomous multi-agent planner are outside this release. Check the remote agent?s protocol support before enabling a connection.
Keep permissions specific to each tenant and agent
Agent and tool registries belong to the tenant that created them. Administrators choose which published agents a grant can invoke, set an expiry, and revoke access when it is no longer needed. Upstream credentials are stored encrypted. Outbound calls use restricted HTTPS transport, while request limits and invocation records help administrators investigate failures and unexpected usage.
Configure publishing for ChatGPT and other MCP clients
The publishing flow supports manually registered OAuth public clients, exact redirect URIs, PKCE S256, and explicit administrator consent. Authorization codes are short lived and single use; access tokens are resource bound and expire after one hour. Refresh tokens and dynamic client registration are not supported in this release. Client compatibility and any marketplace or directory approval must be verified separately; publishing a configuration does not automatically list an agent in ChatGPT.
Prepare a connection before enabling access
Start with one agent and one clearly defined question. Review the agent?s knowledge sources, register the remote endpoint or publishing client, and grant the minimum access required. Inspect the response and activity record before expanding access to additional tools or agents. The developer documentation and downloadable interoperability guide describe endpoint paths, supported protocol operations, and authentication limits.