# AI Agent Client Compatibility > Use the current compatibility matrix and read-only starter configurations, then record live conformance against an exact client build or review date before making a support claim. - Human page: https://mailrith.com/developers/agent-compatibility - Markdown page: https://mailrith.com/developers/agent-compatibility.md - Category: AI Agent Platform - Reading time: 8 min read - Last updated: 2026-07-22 - Related keywords: AI Agent Client Compatibility, AI Agent Client Compatibility developer docs, AI Agent Platform, AI Agent Platform developer docs, Mailrith developer docs, Mailrith public API, Supported Client Matrix, Copyable Configurations, Known Limitations, Conformance And Support, MCP Server, AI Agent Architecture And Security, AI Agent Examples ## AI Agent Notes - Use this page as implementation guidance, then validate exact endpoint fields against the OpenAPI document. - Keep API keys server-side and workspace-scoped unless a guide explicitly says otherwise. - Do not invent privacy, consent, or lawful-basis evidence. Send only fields that appear in the OpenAPI schema for the endpoint you are using. ## What this guide covers Check reviewed OpenAI, Claude, Codex, n8n, and Pipedream connection shapes, task-specific configurations, known limitations, and conformance rules. ## Supported Client Matrix Mailrith supports exact conformance-reviewed connection shapes rather than claiming every past and future client release. Rolling client evidence expires after 90 days or immediately after a material MCP or authorization change. - OpenAI Responses API: remote MCP, Bearer authorization, allowed-tool filtering, and provider approval controls. Template reviewed July 22, 2026. - Claude Messages API: `mcp-client-2025-11-20`, remote server authorization, and MCP toolset allowlisting. Template reviewed July 22, 2026. - Codex CLI and app: Streamable HTTP MCP, OAuth login or a Bearer-token environment variable, enabled-tool allowlisting, and write approvals. Template reviewed July 22, 2026. - n8n: inactive starter workflow using HTTP Request node 4.3, secret-backed Bearer authentication, and a bounded capability read. - Pipedream: Node.js component starter with a secret property, bounded capability read, and no write side effect. ## Copyable Configurations The versioned connector templates ship with the Mailrith Agent Skill package. Each template contains unresolved secret placeholders and starts read-only or disabled. Resolve secrets in the client at runtime and expand tool access only after reviewing the task. **Remote MCP Connection** ```text Name: Mailrith Transport: Streamable HTTP URL: https://api.mailrith.com/mcp Authentication: OAuth (preferred) or Bearer token from a secret store Initial tools: discovery_get_capabilities, subscribers_list Provider approval: Always ``` **Task-Specific Toolsets** ```text Read-only: discovery_get_capabilities, subscribers_list, broadcasts_list, broadcasts_get Drafting: read-only tools plus broadcasts_create, broadcasts_update, broadcasts_preflight Approved send: drafting tools plus agent_actions_get, agent_actions_issue_approval_token, broadcasts_send Never expose approval tokens to the model or persist them. ``` ## Known Limitations The npm, PyPI, CLI, MCP, and Agent Skill prerelease artifacts are source-validated but are not publicly installable until the coordinated registry workflow completes. Public docs must not claim a registry version is available before the release manifest is marked published. OpenAI, Claude, Codex, n8n, and Pipedream change independently. Re-run conformance after client authentication, MCP, tool approval, or workflow export fields change. n8n and Pipedream starters use stable HTTP primitives for the first read because third-party MCP node formats may change independently. Add writes only with task-specific credentials and Mailrith approval controls. Agent activity is deliberately redacted and retained for a bounded period. It is an operational audit trail, not a copy of prompts, Subscriber records, or message content. Mailrith cannot verify claims generated by a model or replace the sender's legal, consent, deliverability, or provider responsibilities. ## Conformance And Support Repository conformance validates discovery, authentication, tool listing, a bounded read, draft behavior, denial of an unapproved high-impact action, approval and execution state, activity correlation, and connector-template safety. Live client runs must record the exact client build or review date and must use a sandbox workspace for writes. > Live OpenAI, Claude, Codex, n8n, and Pipedream conformance remains a release gate until the results are recorded against deployed endpoints. **Repository Conformance** ```bash pnpm agent:clients:conformance -- --profile static pnpm agent:clients:conformance -- --profile read --client codex --json ``` ## Related Guides - [MCP Server](https://mailrith.com/developers/mcp.md): Mailrith's official MCP server is the highest-level interface for agent runtimes. This page covers the remote endpoint and includes copy-paste examples for OpenAI, Claude, n8n, and Pipedream. - [AI Agent Architecture And Security](https://mailrith.com/developers/agent-security.md): Mailrith uses one public resource model, least-privilege credentials, typed operations, exact action approval, existing delivery safeguards, and compact redacted activity across every supported agent client. - [AI Agent Examples](https://mailrith.com/developers/agent-examples.md): Each example demonstrates one safe Mailrith workflow, lists its required permissions, and keeps reads, retries, writes, approvals, and stored data bounded.