Functional Intelligence: Agents, MCP and maybe the end of the Web?

We spend countless hours navigating poorly designed interfaces, searching for information, and completing tedious tasks like digital form filling—fighting with systems that should serve us, not consume our time. What if we could eliminate these frustrating interfaces and processes entirely?

This talk touches on a future where autonomous systems, powered by standardised protocols, transform how we interact with digital services and fundamentally reshape the human web experience. Instead of humans adapting to clunky UIs and processes, intelligent agents handle these tasks by directly communicating with each other.

The Model Context Protocol (MCP) is leading this transformation by providing a standardised way to provide AI models with contextual information, allowing them to act on our behalf. Using a real-world example, we'll explore MCP's core primitives and demonstrate how http4k's new MCP SDK fuses these concepts with the functional, clean, and testable APIs that http4k is known for.

You'll see how a task that once required multiple manual steps becomes automated through agentic interfaces built with just a few simple functions—all while maintaining http4k's philosophy of zero magic and exceptional testability.

This isn't just about efficiency—it's about the next wave of technological change that could fundamentally transform our relationship with digital systems, and reshape how we build services that genuinely help our users rather than frustrate them.

Transcript

1. Functional Intelligence: Agents, MCP and the end of the Web(?)
2. Hello, it’s David! https://clockwork.ing/ https://monopolis.cloud GoogleDevExpert
3. What’s the plan? •What have we done to ourselves? •LLMs to the rescue… maybe •Smoothing the path with MCP (and http4k!) •Reimagining how things could be •Some words of warning…
4. “Just fill in this form”* Me *Names have been changed to protect the guilty
5. “Just fill in this form”* Me My Mac 1. Get my details *Names have been changed to protect the guilty
6. “Just fill in this form”* Me My Mac 1. Get my details Acme Insurance.com 2 . Search for the form *Names have been changed to protect the guilty
7. “Just fill in this form”* Me My Mac 1. Get my details Acme Insurance.com 2 . Search for the form *Names have been changed to protect the guilty 3 . Phone the call centre
8. “Just fill in this form”* Me My Mac 1. Get my details Acme Insurance.com 2 . Search for the form 4. Lookup orders Rainforest.com *Names have been changed to protect the guilty 3 . Phone the call centre
9. “Just fill in this form”* Me My Mac 1. Get my details Acme Insurance.com 2 . Search for the form 5. Get invoice 4. Lookup orders Rainforest.com *Names have been changed to protect the guilty 3 . Phone the call centre
10. “Just fill in this form”* Me My Mac 1. Get my details Acme Insurance.com 2 . Search for the form 5. Get invoice 4. Lookup orders Rainforest.com *Names have been changed to protect the guilty 3 . Phone the call centre 6. Convert file to PDF
11. “Just fill in this form”* Me My Mac 1. Get my details Acme Insurance.com 2 . Search for the form 7. Raise claim, attach invoice 5. Get invoice 4. Lookup orders Rainforest.com *Names have been changed to protect the guilty 3 . Phone the call centre 6. Convert file to PDF
12. “Just fill in this form”* Me My Mac 1. Get my details Acme Insurance.com 2 . Search for the form 8. Add a todo to track it 7. Raise claim, attach invoice 5. Get invoice 4. Lookup orders Rainforest.com *Names have been changed to protect the guilty 3 . Phone the call centre 6. Convert file to PDF
13. The waste here is legion! •Missing/impenetrable instructions •Dumb chatbots/call centres •Bad UIs and user hostile forms •Manual data gruntwork
14. The waste here is legion! •Missing/impenetrable instructions •Dumb chatbots/call centres •Bad UIs and user hostile forms •Manual data gruntwork
15. It’s ok though. AI will save us!
16. LLMs are very clever!
17. LLMs are also very stupid!
18. LLMs are also very stupid!
19. LLMs are also very stupid!
20. LLMs are also very stupid!
21. How do LLM interactions work? LLM Client Soon-to-be automated worker
22. How do LLM interactions work? LLM Client Soon-to-be automated worker Question ModelParameters Message Message Message Tools Question
23. How do LLM interactions work? LLM Client Soon-to-be automated worker Question ModelParameters Message Message Message Tools Question Message Content Usage Display Content
24. How do LLM interactions work? LLM Client Soon-to-be automated worker Question ModelParameters Message Message Message Tools Question Message Content Usage Display Content Message Message Message Message ModelParameters Message Message Message Tools Follow up
25. How do LLM interactions work? LLM Client Soon-to-be automated worker Question ModelParameters Message Message Message Tools Question Message Content Usage Display Content Message Message Message Message ModelParameters Message Message Message Tools Follow up Message Message Message Content Usage More content
26. Conclusions •LLMs are logical and can visualise contextually •But they are too naive to be trusted •To be truly useful, they need to be connected to: 1. Real world data 2. Some decent instructions!
27. Model Context Protocol
28. What is MCP? •Defines a common unified API for clients to connect •Standardises message format •Reimagines interaction model as capabilities “Enables seamless integration between LLM applications and external data sources and tools”
29. Credit: Norah Sakal @norahsakal
30. There is a type bit of hype! •Number of servers 6494* •SDKs in Go, Java, TS, Python…. •Rapid adoption by big-tech - the below were announced whilst I wrote this talk *At time of slide
31. http4k-mcp-desktop Secure, multi-OS protocol bridge MCP SDK Build functional & testable MCP servers MCP Client Connect to MCPs (HTTP, SSE, WS)
32. MCP Essentials •Standardised message format for interactions (JSON-RPC) •Reimagines interaction model as capabilities: Completions Prompts Tools Resources Sampling
33. Templated instructions for an LLM Prompts List Prompts Get prompt for {arguments} Prompt list updated notification Client Server
34. Prompts Named Prompt + Typesafe Args →Prompt Content
35. Complete for {input} Completions Provide autocompleted text Client Server
36. Completions Convert text argument → Text completed options
37. List Tools Call tool {arguments} Tool list updated notification Tools Perform an (user authorised) structured interaction Client Server
38. Tools Named Tool + Typesafe Args →Content
39. List Resources Read Resource {url} Resource list updated notification Resources Retrieve/subscribe to URI-based data Subscribe {url} Resource {url} updated notification Client Server
40. Resources Convert a URL → Content
41. Unit testing: Capabilities In-memory → super fast!
42. MCP “Server” (Local) MCP Setup: 2024 spec Workstation LLM Client MCP “Server” (Bridge) HTTP LLM Cloud Service HTTP StdIo Meat-based Supervisor Remote Remote (API) (LLM API) MCP Server (SSE) SSE (PUSH) IO = File access Remote StdIo IO HTTP (POST)
43. MCP “Server” (Local) MCP Setup: 2025 spec Workstation LLM Client HTTP LLM Cloud Service HTTP Remote Remote (API) (LLM API) MCP Server (HTTP) = File access Remote IO HTTP (+optional SSE) StdIo Meat-based Supervisor StdIo MCP “Server” (Bridge) IO
44. an entire HTTP ecosystem • Support for 2025 MCP spec in > v6.6.0.0 • Full integration with http4k Security APIs • Zero reflection == Serverless & GraalVM • Pro license or as part of http4k EE • Free for non-commercial, non-profit, research https://mcp.http4k.org
45. Putting it all together Protocol + Identity + Capabilities + Backend → MCP Server
46. Now available in Serverless Protocol + Identity + Capabilities + Adapter = MCP Serverless
47. In-memory → super fast! Unit testing: MCP Servers
48. Remote testing: MCP Servers
49. Let’s revisit our process LLM (supervised by me) *Names have been changed to protect the guilty
50. Let’s revisit our process LLM (supervised by me) My Mac 1. Completion (Details) *Names have been changed to protect the guilty
51. Let’s revisit our process LLM (supervised by me) My Mac 1. Completion (Details) Acme Insurance.com 2 . Prompt (instructions) *Names have been changed to protect the guilty
52. Let’s revisit our process LLM (supervised by me) My Mac 1. Completion (Details) Acme Insurance.com 2 . Prompt (instructions) *Names have been changed to protect the guilty 3 . Have a nice cup of tea
53. Let’s revisit our process LLM (supervised by me) My Mac 1. Completion (Details) Acme Insurance.com 2 . Prompt (instructions) 4. Read Resource (Orders) Rainforest.com *Names have been changed to protect the guilty 3 . Have a nice cup of tea
54. Let’s revisit our process LLM (supervised by me) My Mac 1. Completion (Details) Acme Insurance.com 2 . Prompt (instructions) 5. Tool Call (Invoice) 4. Read Resource (Orders) Rainforest.com *Names have been changed to protect the guilty 3 . Have a nice cup of tea
55. Let’s revisit our process LLM (supervised by me) My Mac 1. Completion (Details) Acme Insurance.com 2 . Prompt (instructions) 5. Tool Call (Invoice) 4. Read Resource (Orders) Rainforest.com *Names have been changed to protect the guilty 6. Tool call (Convert) 3 . Have a nice cup of tea
56. Let’s revisit our process LLM (supervised by me) My Mac 1. Completion (Details) Acme Insurance.com 2 . Prompt (instructions) 7. Tool call (Raise) 5. Tool Call (Invoice) 4. Read Resource (Orders) Rainforest.com *Names have been changed to protect the guilty 6. Tool call (Convert) 3 . Have a nice cup of tea
57. Let’s revisit our process LLM (supervised by me) My Mac 1. Completion (Details) Acme Insurance.com 2 . Prompt (instructions) 8. Tool call (Track) 7. Tool call (Raise) 5. Tool Call (Invoice) 4. Read Resource (Orders) Rainforest.com *Names have been changed to protect the guilty 6. Tool call (Convert) 3 . Have a nice cup of tea
58. There are 6494* servers to install… *At time of slide …. so get going! ! Client MCPs
59. NOT SO FAST…
60. NOT SO FAST…
61. NOT SO FAST…
62. NOT SO FAST…
63. NOT SO FAST…
64. How does Tool Calling work? LLM Client MCP
65. How does Tool Calling work? LLM Client MCP ModelParameters Message Message Message Tools Message
66. How does Tool Calling work? LLM Client MCP ToolCall Usage ModelParameters Message Message Message Tools Message
67. How does Tool Calling work? LLM Client MCP ToolCall Usage ToolCall Approve ModelParameters Message Message Message Tools Message
68. How does Tool Calling work? LLM Client MCP ToolCall Usage Result Message Message Result ModelParameters Message Message Message Tools ToolCall Approve ModelParameters Message Message Message Tools Message
69. How does Tool Calling work? LLM Client Usage Message Message Message Content MCP ToolCall Usage Result Message Message Result ModelParameters Message Message Message Tools ToolCall Approve ModelParameters Message Message Message Tools Message
70. We can create EvilMCP™ The LLM treats tools as “trusted input”
71. Tool poisoning attack Client O!cial MCP EvilMCP™ LLM Query
72. Tool poisoning attack Client O!cial MCP EvilMCP™ LLM Nice tool Query
73. Tool poisoning attack Client O!cial MCP EvilMCP™ LLM Poison tool Poisons intercepts operation Nice tool Query
74. Tool poisoning attack Client O!cial MCP EvilMCP™ LLM Poison tool Poisons intercepts operation Nice tool Approves usage Query
75. Tool poisoning attack Client O!cial MCP EvilMCP™ LLM Poison tool Poisons intercepts operation Nice tool Approves usage Query with poisoned tool instructions LLM requests poisoned action Query
76. Tool poisoning attack Client O!cial MCP EvilMCP™ LLM Poison tool Poisons intercepts operation Nice tool Approves usage Query with poisoned tool instructions LLM requests poisoned action Query Call with exfiltration
77. Recommendations •Exercise extreme caution! •Access to your machine == GAME OVER •95% of all published servers are “uno!cial” •MCP currently safe for purely internal tools •Lots of work for community to do!
78. Summary: http4k is still bullish on MCP •MCPs give LLMs superpowers… •…but open up a new world of risk •LLM visualisations could replace UIs •Expect ecosystem gold rush … with caveats
79. Thank you! Questions? David Denton Engineering Lead | CICD | Security | Advisory | Training https://clockwork.ing/ https://mcp.http4k.org