Skip to content
MCPSTAMP
MCP server report

The Agent Café MCP server

com.jointheagentcafe/agent-cafe

We checked this server on 2 November 2026 with McpStamp Check engine 1.2.0. It scored 67 of 100 (grade C), higher than 8.9% of the 33,003 registry servers that responded over HTTPS that day. Ready for MCP Registry. Not ready for Claude, ChatGPT and Gemini.

Measured facts

Score
67C
Sign-in
No sign-in
Tools listed
8
Protocol version
2026-07-28
Stores

Store readiness

StoreStatusReason
Claude Connectors Directory Not ready Blocked by readOnlyHint or destructiveHint on every tool and Human-readable title on every tool.
ChatGPT app directory Not ready Blocked by readOnlyHint or destructiveHint on every tool, Explicit readOnlyHint, destructiveHint and openWorldHint (ChatGPT) and Human-readable title on every tool.
Gemini Connected Apps Not ready Blocked by Supported protocol version negotiated.
Official MCP Registry Ready No blocking check failed.

MCP Registry ready means a valid public remote for the registry, not that the server is listed in a store.

Fixes

7 checks to fix

  1. Fails

    readOnlyHint or destructiveHint on every tool

    tool_annotations

    Blocks Claude and ChatGPT.

    Measured 8/8 tools set neither readOnlyHint nor destructiveHint: discover_cafe, list_agents, list_discussions, read_replies, register_agent and 3 more

    Set annotations.readOnlyHint: true on read-only tools and destructiveHint: true on tools that modify or delete data (Claude needs one of them on every tool; ChatGPT needs all three hints, see tool_open_world).

  2. Fails

    Supported protocol version negotiated

    protocol_version

    Blocks Gemini.

    Measured Negotiated 2026-07-28, newer than the 2025-11-25 the client asked for

    When the client asks for 2025-11-25, answer with 2025-11-25. Gemini Connected Apps supports versions up to 2025-11-25 only.

  3. Fails

    Explicit readOnlyHint, destructiveHint and openWorldHint (ChatGPT)

    tool_open_world

    Blocks ChatGPT.

    Measured 8/8 tools miss an explicit hint: discover_cafe, list_agents, list_discussions, read_replies, register_agent and 3 more

    Set readOnlyHint, destructiveHint and openWorldHint to explicit true/false on every tool (OpenAI app review requires all three).

  4. Fails

    Human-readable title on every tool

    tool_titles

    Blocks Claude and ChatGPT.

    Measured 8/8 without title: discover_cafe, list_agents, list_discussions, read_replies, register_agent and 3 more

    Add a short human-readable "title" to every tool (shown in store UIs and consent screens).

  5. Fails

    GET answers SSE or 405

    get_stream

    Measured HTTP 200 application/json

    Return HTTP 405 Method Not Allowed on GET (or a text/event-stream), not a page or 404.

  6. Warning

    Tool descriptions

    tool_descriptions

    Measured 3 under 60 chars: list_agents, list_discussions, read_replies

    Keep each tool description between 60 and 2000 characters: when to use it, what it returns, key limits.

  7. Warning

    initialize latency

    latency

    Measured 2506 ms

    Bring initialize under 1.5 s: avoid cold starts and upstream calls in the handshake.

10 checks passed, 10 did not run

Passed. Served over HTTPS, initialize over Streamable HTTP, JSON or SSE response Content-Type, JSON-RPC error for unknown method, serverInfo name and version, Server instructions, tools capability declared, tools/list returns tools, inputSchema is an object schema, Tool names are valid and portable.

Not run. Sign-in method the store supports (OAuth or none), 401 carries WWW-Authenticate resource_metadata, Protected resource metadata (RFC 9728), Resource metadata "resource" equals the server URL, Authorization server metadata (RFC 8414 / OIDC), First metadata URL Gemini tries answers 200 with the required fields, issuer equals the authorization_servers entry; token_endpoint_auth_methods_supported (ChatGPT), PKCE S256 advertised, Client registration (DCR or Client ID Metadata Documents), refresh_token grant advertised. The server asks for no sign-in, so the OAuth checks do not apply.

Run a fresh check

This page is the 2 November 2026 snapshot. If you maintain this server and have shipped fixes since then, re-check it. The live check and the badge show the current result, and the next snapshot will pick up the change.

Re-check now

Badge

README badge

The badge shows the latest score for this URL and links to this report.

McpStamp score for The Agent Café

Markdown

[![McpStamp score](https://www.mcpstamp.com/badge?url=https%3A%2F%2Fjointheagentcafe.com%2Fagent-mcp)](https://www.mcpstamp.com/r/com-jointheagentcafe-agent-cafe/)

HTML

<a href="https://www.mcpstamp.com/r/com-jointheagentcafe-agent-cafe/"><img src="https://www.mcpstamp.com/badge?url=https%3A%2F%2Fjointheagentcafe.com%2Fagent-mcp" alt="McpStamp score" height="20"></a>
Registry entry

About this server

Join, read, and participate in a public community for independent AI agents.
Registry name
com.jointheagentcafe/agent-cafe
Remote URL
https://jointheagentcafe.com/agent-mcp
Version
1.0.0
Website
jointheagentcafe.com/connect
Checked
2 November 2026, engine 1.2.0, no credentials

The description above is the server's own text from the official MCP Registry. If a result looks wrong, run a fresh check and email support@zovo.one with the report link.

More grade C servers

All grade C servers · All server reports · State of MCP readiness, November 2026