The hosted MCP endpoint is https://mcp.converterhq.io/mcp. It is a stateless Streamable HTTP endpoint backed by the same ConverterHQ resources as the REST API. Delegated HQTools OAuth keeps permanent API keys out of AI clients.
Connect
Add https://mcp.converterhq.io/mcp as a Streamable HTTP server. ConverterHQ exposes OAuth protected-resource metadata, supports server/discover, and currently supports the2026-07-28 MCP protocol. Every request must send that same version in the MCP-Protocol-Version header and inparams._meta["io.modelcontextprotocol/protocolVersion"], together with a client-capabilities object. Older initialize-style clients are not part of the supported launch matrix.
POST /mcp HTTP/1.1
Host: mcp.converterhq.io
MCP-Protocol-Version: 2026-07-28
Mcp-Method: server/discoverOAuth and scopes
The MCP endpoint accepts delegated OAuth access tokens, not permanent ConverterHQ API keys. Request only the scopes the agent needs: capability discovery, Job/Task operations, Workflows and Usage are independently authorized.
Tool catalog
ConverterHQ exposes 13 tools: convert_file, estimate_conversion, list_formats, get_conversion_capability, list_conversion_capabilities, create_job, validate_job, get_job, cancel_job, get_task, list_workflows, run_workflow and get_usage. Capability tools are safe for planning; execution tools create the same durable resources visible in REST and the developer workspace.
Durable Tasks
Clients negotiating the MCP Tasks extension can receive a durable MCP task mapped to a ConverterHQ Job and later call the Tasks methods to inspect or cancel it. Clients without the extension receive the Job handle and can continue with get_job. The original tool call does not need to remain connected for conversion execution to finish.
Workflow variables
run_workflow accepts the same validated variable map, client_reference_id and metadata as POST /v1/workflows/{id}/run. Variable schemas constrain what an agent can substitute before the Job is created.
Operating safely
Use least-privilege OAuth scopes, keep human approval in front of sensitive external side effects where your agent requires it, persist returned Job IDs, and use authoritative REST/Job state for support and recovery. Revoke the OAuth grant when the client is no longer trusted.