UTCP (Universal Tool Calling Protocol) is an open standard for AI agents to call tools directly, positioned as a lightweight alternative to MCP. It eliminates the "wrapper tax" for lower latency by using a simple JSON manifest to connect to native APIs.
Hi everyone!
In the world of AI agents, the Model Context Protocol (MCP) has quickly become the dominant, industry-backed standard for tool calling. But a new, community-driven challenger has emerged with a completely different philosophy: the Universal Tool Calling Protocol (UTCP).
UTCP's approach is all about direct connection. Instead of requiring a wrapper server like MCP, it lets an agent call a tool's native endpoint (like a REST API) directly. This means lower latency, no "wrapper tax," and you can reuse all your existing security and auth.
While UTCP is still in its early days and doesn't have MCP's massive ecosystem backing, I don't expect it to challenge that dominance overnight. But sometimes the opposite of a good strategy is also a good strategy. I believe UTCP has the potential to grow a new ecosystem. After all, change, innovation, and competition are what ultimately make everything better.
Report
Without wrapper-driven standardization, how do you plan to prevent integration fragmentation that could hurt interoperability between tools?
@vouchy we believe that communication transport layers have already been standardized enough (e.g, HTTP, gRPC), so we can just leverage the existing ones instead of creating a new transport standard (MCP)
In a nutshell, UTCP leverages the interoperability of current standards, while MCP requires an ecosystem-wide rebuild :)
@juan_viera_garcia1@aliraza36580 why not move your UTCP discovery endpoint to `/.well-known` since .well-known URI path is a special directory defined by the IETF in RFC 5785. It allows web services to expose information or configuration files in a predictable, standardized location.
@juan_viera_garcia1 highly unlikely unless you 1) have a load of funding or 2) get one of the major labs to support it.
That's how we got OAuth adopted — a bunch of scrappy upstarts got together to define the protocol and then once Google came onboard, that gave it the legitimacy and visibility it needed to take off.
@chrismessina read how DeWitt Clinton came aboard the OAuth project, and then you developed it very closely with the Google team. Was this something planned from your side, or serendipity? And if you could draw parallels, what could be a good strategy to go down the same path with UTCP?
@aliraza36580 primarily serendipity... just as I did with the hashtag, I did a lot of ground work to lay the foundation by soliciting lots of feedback, input, and trial and error improvements so that by the time Google came along (and to be fair Dewitt was already active in the OSS community), we'd already established our goals, had rough consensus and running code.
We also took care to license OAuth using a framework that could be adopted by the major platforms: the OWFa 1.0 Agreement.
@chrismessina thanks for sharing Chris! We're definitely focusing on the groundwork and working very closely with the community to shape UTCP, let's see where serendipity takes it.
This sounds like a solid move toward more open, efficient tool integration for AI agents. Love the focus on reducing wrapper overhead. Looking forward to trying it out when I can. Congrats on the launch 🚀
Appreciate the fresh perspective UTCP brings with direct API calls. How do you see it handling security concerns compared to MCP’s wrapper server approach? @aliraza36580
@rachitmagon Thanks! UTCP leverages the tool’s native, battle-tested security and is thus able to call any tool while guaranteeing the same security as if the tool was called by a human.
MCP, on the other hand, introduces a security tax by requiring authentication and authorization to be re-implemented and managed at the MCP layer. This is not scalable and introduces a lot of new attack vectors.
UTCP.io seems like an elegant solution love how it centralizes team communications with secure, customizable workspaces. Quick question are you planning to add integrations with popular project management tools like Jira or Trello to streamline workflows even further?
Replies
Flowtica Scribe
Without wrapper-driven standardization, how do you plan to prevent integration fragmentation that could hurt interoperability between tools?
UTCP
@vouchy we believe that communication transport layers have already been standardized enough (e.g, HTTP, gRPC), so we can just leverage the existing ones instead of creating a new transport standard (MCP)
In a nutshell, UTCP leverages the interoperability of current standards, while MCP requires an ecosystem-wide rebuild :)
AltPage.ai
Cutting out that "wrapper tax" with just a simple JSON manifest? That’s honestly genius, fr. Been waiting for something like this—props to the team!
Raycast
@juan_viera_garcia1 @aliraza36580 why not move your UTCP discovery endpoint to `/.well-known` since .well-known URI path is a special directory defined by the IETF in RFC 5785. It allows web services to expose information or configuration files in a predictable, standardized location.
UTCP
@chrismessina Chris that’s a great idea actually! Appreciate it
UTCP
@chrismessina created a ticket (link) 💪
UTCP
What do you think, is it possible to create a popular protocol that rivals MCP?
Raycast
@juan_viera_garcia1 highly unlikely unless you 1) have a load of funding or 2) get one of the major labs to support it.
That's how we got OAuth adopted — a bunch of scrappy upstarts got together to define the protocol and then once Google came onboard, that gave it the legitimacy and visibility it needed to take off.
UTCP
@chrismessina Hi Chris, totally agree
That's why we're proactively reaching out to the major labs to find a collaboration together
However, if you know anyone that might be a good lead, would be great to connect!
UTCP
@chrismessina read how DeWitt Clinton came aboard the OAuth project, and then you developed it very closely with the Google team. Was this something planned from your side, or serendipity? And if you could draw parallels, what could be a good strategy to go down the same path with UTCP?
Raycast
@aliraza36580 primarily serendipity... just as I did with the hashtag, I did a lot of ground work to lay the foundation by soliciting lots of feedback, input, and trial and error improvements so that by the time Google came along (and to be fair Dewitt was already active in the OSS community), we'd already established our goals, had rough consensus and running code.
We also took care to license OAuth using a framework that could be adopted by the major platforms: the OWFa 1.0 Agreement.
UTCP
@chrismessina thanks for sharing Chris! We're definitely focusing on the groundwork and working very closely with the community to shape UTCP, let's see where serendipity takes it.
Linkinize
This sounds like a solid move toward more open, efficient tool integration for AI agents. Love the focus on reducing wrapper overhead. Looking forward to trying it out when I can. Congrats on the launch 🚀
Smoopit
Appreciate the fresh perspective UTCP brings with direct API calls. How do you see it handling security concerns compared to MCP’s wrapper server approach? @aliraza36580
UTCP
@rachitmagon Thanks! UTCP leverages the tool’s native, battle-tested security and is thus able to call any tool while guaranteeing the same security as if the tool was called by a human.
MCP, on the other hand, introduces a security tax by requiring authentication and authorization to be re-implemented and managed at the MCP layer. This is not scalable and introduces a lot of new attack vectors.
Smoopit
@aliraza36580 Got it, appreciate the breakdown.
Kandid
UTCP.io seems like an elegant solution love how it centralizes team communications with secure, customizable workspaces. Quick question are you planning to add integrations with popular project management tools like Jira or Trello to streamline workflows even further?
UTCP
@pulkitgarg Added it as a ticket for us! https://github.com/universal-tool-calling-protocol/utcp-specification/issues/14
Nas.io
Open API alternatives are critical. What’s your approach to keeping the tool-calling secure?
This is amazing! I never understood why MCP as a protocol needs an extra server when your actual server already has an API layer.