🚨 Biggest update to MCPs yet! 🚨 It is severely understated how many MCPs fail to connect to different hosts because of confusing descriptions and schemas. The core issue is that they are confusing for their actual users, those being your Codexes and Cursors. And MCP firms themselves cannot check it because they simply can't see what is happening inside of the client. On the other side of this relationship, the client can drop your MCP and use another one due to reasons like a description that disagrees with the schema or paired locators required only in the prose. So we added "Contracts" to your @AgentStatus profile, where you can see all the issues the host encounters when it connects to your MCP and prevent them! Saving yourself money and peace of mind 😌 P.s. If you are still not listed on agentstatus.dev/mcp-index what are you doin??

I wish more people saw this post. @X algorithm, please give it a chance

I had so much to say about MCPs that I just decided to record a quick 40 sec video instead Make sure to check agentstatus.dev/mcp-index and engage 😌

if you have an MCP that is still not listed on MCP Index, you should be locked up for criminal activity 👮‍♂️ agentstatus.dev/mcp-index

what's the coolest MCP have you came across?

Hey there! We built an #MCP Index, which is an independent ranking of MCP servers by quality and popularity. agentstatus.dev/mcp-index All MCPs are graded with a letter grade that is calculated based on things like reach, speed, tools and spec. There is an entire methodology you can look into. The goal of the project is to create unified and fully independent list that verifies how MCPs perform with every major host, what tools it offers, if their descriptions align with their schemas and so on. I think its also a great way for anyone to showcase their MCP. You can add it to the ranking for free so it gets independently verified and scored (and you see why you get a certain grade/position). You can also get a badge on your website (kinda like featured on product hunt). You can also then view how your MCP is being accessed by different hosts from residential devices. But if you are not building but rather using MCPs, this is intended to be an extended, fully independent library. I please ask you to check it out and, if you like it, engage with it. If you have questions or recommendations how I can make it better, I will gladly accept them as I still work on this and am trying to make it simple, yet comprehensive. Thank you and I am looking forward to hearing your feedback!!

how does insurance know that your ai agents won't bust and harm someone's system?

Is it just me or twitter is like 80% bots now? My last video had at least 5 of them and not a single real human. Lowkey, so concerned

View the way your MCP server works from different hosts. Experience your MCP the way the user experiences it from their Claude Code or Cursor. Try at @AgentStatus

Today @AgentStatus we're featuring @prisma. In the latest Prisma run on AgentStatus, 16 of 16 tested hosts including passed the compatibility checks, including reaching the server, authorization, starting an MCP session, and sending the required headers. Congratulations to the team!!! @prisma @sorenbs @sneub @william_ma14406

Have you wondered how most popular MCPs hold up against each other in their quality? Well, we at @AgentStatus definitely did. So we added a popularity ranking that has a summary of all calls that are sent to MCPs in our MCP Index. To tease it, here is a small comparison of Reachability of 5 of them out of the top 15 most popular MCPs. We dig a lot deaper than just reachability and end up grading them all based on their reach, tools, speed and spec. @theagenttimes_ @NLM_NIH @OOBEonSol @supabase View the whole quality comparison at agentstatus.dev/mcp-index Reach out if you'd like your MCP to be featured next

Looking to #connect with founders building cool stuff! Especially ai agents, mcps and you continue the list;)

Today @AgentStatus we're highlighting Marcopolo(marcopolo.dev/) MarcoPolo's MCP reached the server on 16 of 16 hosts. All sixteen passed authorization, initialized an MCP session, and sent the required headers. For the two authorized host profiles used for tool testing, both completed their selected tool calls, and no host produced a confirmed MarcoPolo server failure. Congratulations to the team!!

Decided to record a silly video to explain you why @AgentStatus added Host Compatibility tab to their dashboard. @attio thank you for running a great business with over 40 tools in your MCP! See how your MCP shows up in different hosts here --> agentstatus.dev/

Today at @AgentStatus, we're spotlighting @attio @nicolasosharp @byteofbits @mulholo On 16 Sep 2026 at 08:08 UTC, Attio's MCP completed a tool invoke on fourteen of fourteen tested hosts. Congratulations to the team!!!

Next up on @AgentStatus is @semgrep 5 planned checks on @semgrep 's MCP were called, answered, and passed the functional check. semgrep_scan_remote still scored did the job, but the reply was a deprecation notice that hosted scans no longer run. The three counts on the card are the same five calls and here is their meaning: Planned: we meant to invoke them from a residential node and the device sent the tool invoke. Answered: the server returned a complete JSON-RPC result. Did the job: the payload was the tool's data instead of an application error, and it matched the output schema the tool published. No published schema is neutral. We do not fail the server for our time cap or for arguments we sent badly. A residential node opened the hosted @semgrep MCP as @AgentStatus. This is live MCP traffic from a home internet path. get_abstract_syntax_tree, get_supported_languages, semgrep_scan_with_custom_rule, and semgrep_whoami came back and passed those checks. semgrep_scan_remote passed them too because the deprecation notice was complete, usable content. Tool states that hosted scans are gone, run semgrep mcp locally, and findings already on the AppSec Platform stay on semgrep_findings. Independent residential run is complete. Create an account to see the full log and what to fix at agentstatus.dev Congratulations to @semgrep team! @jaysondelancey @0xine @DaghanAltas @erikbuchanan

We independently measured host compatibility for @jamdotdev's public MCP on 15 Sep 2026. @jamdotdev MCP is how agents pull recordings, logs, and transcripts from a bug report, and it completed a tool invoke on 16 of 16 hosts we tested. @AgentStatus does not score from a single ping. Every six hours @AgentStatus nodes open the public MCP URL, reach the server, negotiate the dialect, list the catalog, then call tools and score whether the result looked like the job. We repeat that walk inside each AI host. On this run the catalog listed 33 tools, and 22 of 33 have call history [can be viewed on the dashboard]. The latest connection completed 6 valid standalone calls with usable responses, including listRecordingLinks and listRecordingUrls. Median wall time from connect through a tool result was 3.36s. Congratulations to @jamdotdev and their team. @majroth @kippe07 @matthewrubright @maja_fili @_cichocinski @matheeuspoleza Cards below.

We launched host compatibility for MCPs at @AgentStatus this weekend. It measures whether the MCP server completes a tool invoke on the AI hosts people actually open. It is scored from outside the vendor and outside each host. @agentmail is the first server on the AgentStatus MCP Index with that number - 18 of 18 hosts on 14 Sep 2026. Here is how the number is built. AgentStatus runs a node on a residential network and every six hours that node acts as an MCP client against the public URL. It has to reach the endpoint. Then negotiate the MCP dialect, so both sides agree how to speak. Then list the tools the server advertises. Then actually call one, with a safe read. Then we score the payload such as did inboxes, threads, or messages come back as data that looked like the job, or did the call error and return nothing useful. We walk that same path through 18 hosts, @cursor_ai , @claudeai , @ChatGPT and the rest, because each host is its own client path onto the same server and might have different interaction with a given MCP, or in fact, no interaction at all. The score is independent of @agentmail's logs and of each host's dashboard. Median wall time from connect through tool result was 4.42s, across the hosts that returned clean timing. Congratulations to @agentmail. Run times below.

Today @AgentStatus starts publishing the MCP index. Every 6 hours we take a public MCP server and try to use it from 18 AI hosts such as @cursor_ai , @claudeai , @ChatGPT , @Copilot , @n8n_io and many more, from residential devices. First server in the index: @agentmail So today, we'll be spotlighting @agentmail. On 14 Sep 2026 at 09:34 UTC, @agentmail 's MCP completed a tool invoke on eighteen of eighteen hosts. There's currently no other way for teams to track how their MCP behaves across AI hosts from residential devices Shoutout to Agentmail team! @binoyperera @sidharth6 @michaelhyunkim @haakamaujla @sanjith_sambath

hey. let this tweet find people who work with mcp. im looking for you guys