1(globalThis.webpackChunkmy_website=globalThis.webpackChunkmy_website||[]).push([[11456],{67883(e,i,l){e.exports={src:Object.assign(Object.create({toString(){return this.src}}),{srcSet:l.p+"assets/ideal-img/mcp_ui.c382d1d.640.png 640w,"+l.p+"assets/ideal-img/mcp_ui.e81a3cf.1280.png 1280w",images:[{path:l.p+"assets/ideal-img/mcp_ui.c382d1d.640.png",width:640,height:363},{path:l.p+"assets/ideal-img/mcp_ui.e81a3cf.1280.png",width:1280,height:726}],src:l.p+"assets/ideal-img/mcp_ui.c382d1d.640.png",placeholder:void 0,width:640,height:363}),preSrc:"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAoAAAAGCAYAAAD68A/GAAAACXBIWXMAAAsTAAALEwEAmpwYAAAAr0lEQVR4nEWMO04DQRBE92pcgnMQcxwugYggtJDI+ThBBEb+sLtmd3qq2zPPmjUSQUlV3a+q++l7hmFAEu7OKZwajtxxqQLtnrqUEk1ZooQzZOdzisVbzrU/7JmmOXVtpYFt0eTcPm+4vv/gZWdEnuvX+yvj4Q80M8LF5CduHtdc3a142iSQ1X0/cjyOFzAikBxK8LYdeVh/L0VJtf2z2QX8VxA5EfPvkiWvpZRWSGfnJOScQcySfwAAAABJRU5ErkJggg=="}},34832(e,i,l){e.exports={src:Object.assign(Object.create({toString(){return this.src}}),{srcSet:l.p+"assets/ideal-img/ui_usage.47fbd86.640.png 640w,"+l.p+"assets/ideal-img/ui_usage.3ffdba3.1200.png 1200w",images:[{path:l.p+"assets/ideal-img/ui_usage.47fbd86.640.png",width:640,height:334},{path:l.p+"assets/ideal-img/ui_usage.3ffdba3.1200.png",width:1200,height:627}],src:l.p+"assets/ideal-img/ui_usage.47fbd86.640.png",placeholder:void 0,width:640,height:334}),preSrc:"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAoAAAAFCAYAAAB8ZH1oAAAACXBIWXMAAAsTAAALEwEAmpwYAAAAgElEQVR4nGWOuw4CMQwE8/8/wi9xV14TUdwDFOI4tjPoQBSIYqod7W7ato11Xdn3nVIKvfc/3Iy0LAvTNDHPMzlnRIRnrUQMYgwYg9qdZGa4OxGBqtKaME4BPiJwuWbSGX6RplRRRDtH6zwtOB6F272Qfv8YEeAeqMe7sbVzRXkBU1bCBM+B3UkAAAAASUVORK5CYII="}},19912(e,i,l){"use strict";l.r(i),l.d(i,{assets:()=>d,contentTitle:()=>o,default:()=>p,frontMatter:()=>a,metadata:()=>t,toc:()=>h});let t=JSON.parse('{"id":"v1.65.0-stable/index","title":"v1.65.0-stable - Model Context Protocol","description":"v1.65.0-stable is live now. Here are the key highlights of this release:","source":"@site/release_notes/v1.65.0-stable/index.md","sourceDirName":"v1.65.0-stable","slug":"/v1.65.0-stable/v1.65.0-stable","permalink":"/release_notes/v1.65.0-stable/v1.65.0-stable","draft":false,"unlisted":false,"tags":[{"inline":true,"label":"mcp","permalink":"/release_notes/tags/mcp"},{"inline":true,"label":"custom_prompt_management","permalink":"/release_notes/tags/custom-prompt-management"}],"version":"current","frontMatter":{"title":"v1.65.0-stable - Model Context Protocol","slug":"v1.65.0-stable","date":"2025-03-30T10:00:00.000Z","authors":[{"name":"Krrish Dholakia","title":"CEO, LiteLLM","url":"https://www.linkedin.com/in/krish-d/","image_url":"https://media.licdn.com/dms/image/v2/D4D03AQGrlsJ3aqpHmQ/profile-displayphoto-shrink_400_400/B4DZSAzgP7HYAg-/0/1737327772964?e=1749686400&v=beta&t=Hkl3U8Ps0VtvNxX0BNNq24b4dtX5wQaPFp6oiKCIHD8"},{"name":"Ishaan Jaffer","title":"CTO, LiteLLM","url":"https://www.linkedin.com/in/reffajnaahsi/","image_url":"https://pbs.twimg.com/profile_images/1613813310264340481/lz54oEiB_400x400.jpg"}],"tags":["mcp","custom_prompt_management"],"hide_table_of_contents":false},"sidebar":"releaseNotesSidebar","previous":{"title":"v1.65.0","permalink":"/release_notes/v1.65.0/v1.65.0"},"next":{"title":"v1.63.14","permalink":"/release_notes/v1.63.14/v1.63.14-stable"}}');var n=l(74848),s=l(28453),r=l(29815);let a={title:"v1.65.0-stable - Model Context Protocol",slug:"v1.65.0-stable",date:new Date("2025-03-30T10:00:00.000Z"),authors:[{name:"Krrish Dholakia",title:"CEO, LiteLLM",url:"https://www.linkedin.com/in/krish-d/",image_url:"https://media.licdn.com/dms/image/v2/D4D03AQGrlsJ3aqpHmQ/profile-displayphoto-shrink_400_400/B4DZSAzgP7HYAg-/0/1737327772964?e=1749686400&v=beta&t=Hkl3U8Ps0VtvNxX0BNNq24b4dtX5wQaPFp6oiKCIHD8"},{name:"Ishaan Jaffer",title:"CTO, LiteLLM",url:"https://www.linkedin.com/in/reffajnaahsi/",image_url:"https://pbs.twimg.com/profile_images/1613813310264340481/lz54oEiB_400x400.jpg"}],tags:["mcp","custom_prompt_management"],hide_table_of_contents:!1},o,d={},h=[{value:"Model Context Protocol (MCP)",id:"model-context-protocol-mcp",level:2}
1,{value:"UI view total usage after 1M+ logs",id:"ui-view-total-usage-after-1m-logs",level:2},{value:"New Models / Updated Models",id:"new-models--updated-models",level:2},{value:"LLM Translation",id:"llm-translation",level:2},{value:"Spend Tracking Improvements",id:"spend-tracking-improvements",level:2},{value:"UI",id:"ui",level:2},{value:"Model Management",id:"model-management",level:3},{value:"Request Logs",id:"request-logs",level:3},{value:"Usage Tab",id:"usage-tab",level:3},{value:"Logging Integrations",id:"logging-integrations",level:2},{value:"Performance / Reliability Improvements",id:"performance--reliability-improvements",level:2},{value:"General Improvements",id:"general-improvements",level:2},{value:"Security",id:"security",level:2},{value:"Complete Git Diff",id:"complete-git-diff",level:2}];function c(e){let i={a:"a",code:"code",h2:"h2",h3:"h3",li:"li",p:"p",pre:"pre",strong:"strong",ul:"ul",...(0,s.R)(),...e.components};return(0,n.jsxs)(n.Fragment,{children:[(0,n.jsx)(i.p,{children:"v1.65.0-stable is live now. Here are the key highlights of this release:"}),"\n",(0,n.jsxs)(i.ul,{children:["\n",(0,n.jsxs)(i.li,{children:[(0,n.jsx)(i.strong,{children:"MCP Support"}),": Support for adding and using MCP servers on the LiteLLM proxy."]}),"\n",(0,n.jsxs)(i.li,{children:[(0,n.jsx)(i.strong,{children:"UI view total usage after 1M+ logs"}),": You can now view usage analytics after crossing 1M+ logs in DB."]}),"\n"]}),"\n",(0,n.jsx)(i.h2,{id:"model-context-protocol-mcp",children:"Model Context Protocol (MCP)"}),"\n",(0,n.jsxs)(i.p,{children:["This release introduces support for centrally adding MCP servers on LiteLLM. This allows you to add MCP server endpoints and your developers can ",(0,n.jsx)(i.code,{children:"list"})," and ",(0,n.jsx)(i.code,{children:"call"})," MCP tools through LiteLLM."]}),"\n",(0,n.jsxs)(i.p,{children:["Read more about MCP ",(0,n.jsx)(i.a,{href:"https://docs.litellm.ai/docs/mcp",children:"here"}),"."]}),"\n",(0,n.jsx)(r.A,{img:l(67883),style:{width:"100%",display:"block",margin:"2rem auto"}}),"\n",(0,n.jsx)("p",{style:{textAlign:"left",color:"#666"},children:"Expose and use MCP servers through LiteLLM"}),"\n",(0,n.jsx)(i.h2,{id:"ui-view-total-usage-after-1m-logs",children:"UI view total usage after 1M+ logs"}),"\n",(0,n.jsx)(i.p,{children:"This release brings the ability to view total usage analytics even after exceeding 1M+ logs in your database. We've implemented a scalable architecture that stores only aggregate usage data, resulting in significantly more efficient queries and reduced database CPU utilization."}),"\n",(0,n.jsx)(r.A,{img:l(34832),style:{width:"100%",display:"block",margin:"2rem auto"}}),"\n",(0,n.jsx)("p",{style:{textAlign:"left",color:"#666"},children:"View total usage after 1M+ logs"}),"\n",(0,n.jsxs)(i.ul,{children:["\n",(0,n.jsxs)(i.li,{children:["\n",(0,n.jsx)(i.p,{children:"How this works:"}),"\n",(0,n.jsxs)(i.ul,{children:["\n",(0,n.jsx)(i.li,{children:"We now aggregate usage data into a dedicated DailyUserSpend table, significantly reducing query load and CPU usage even beyond 1M+ logs."}),"\n"]}),"\n"]}),"\n",(0,n.jsxs)(i.li,{children:["\n",(0,n.jsx)(i.p,{children:"Daily Spend Breakdown API:"}),"\n",(0,n.jsxs)(i.ul,{children:["\n",(0,n.jsx)(i.li,{children:"Retrieve granular daily usage data (by model, provider, and API key) with a single endpoint.\nExample Request:"}),"\n"]}),"\n",(0,n.jsx)(i.pre,{children:(0,n.jsx)(i.code,{className:"language-shell",metastring:'title="Daily Spend Breakdown API" showLineNumbers',children:"curl -L -X GET 'http://localhost:4000/user/daily/activity?start_date=2025-03-20&end_date=2025-03-27' \\\n-H 'Authorization: Bearer sk-...'\n"})}),"\n",(0,n.jsx)(i.pre,{children:(0,n.jsx)(i.code,{className:"language-json",metastring:'title="Daily Spend Breakdown API Response" showLineNumbers',children:'{\n "results": [\n {\n "date": "2025-03-27",\n "metrics": {\n "spend": 0.0177072,\n "prompt_tokens": 111,\n "completion_tokens": 1711,\n "total_tokens": 1822,\n "api_requests": 11\n },\n "breakdown": {\n "models": {\n "gpt-4o-mini": {\n "spend": 1.095e-05,\n "prompt_tokens": 37,\n "completion_tokens": 9,\n "total_tokens": 46,\n "api_requests": 1\n },\n "providers": { "openai": { ... }, "azure_ai": { ... } },\n "api_keys": { "3126b6eaf1...": { ... } }\n }\n }
1\n ],\n "metadata": {\n "total_spend": 0.7274667,\n "total_prompt_tokens": 280990,\n "total_completion_tokens": 376674,\n "total_api_requests": 14\n }\n}\n'})}),"\n"]}),"\n"]}),"\n",(0,n.jsx)(i.h2,{id:"new-models--updated-models",children:"New Models / Updated Models"}),"\n",(0,n.jsxs)(i.ul,{children:["\n",(0,n.jsxs)(i.li,{children:["Support for Vertex AI gemini-2.0-flash-lite & Google AI Studio gemini-2.0-flash-lite ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9523",children:"PR"})]}),"\n",(0,n.jsxs)(i.li,{children:["Support for Vertex AI Fine-Tuned LLMs ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9542",children:"PR"})]}),"\n",(0,n.jsxs)(i.li,{children:["Nova Canvas image generation support ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9525",children:"PR"})]}),"\n",(0,n.jsxs)(i.li,{children:["OpenAI gpt-4o-transcribe support ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9517",children:"PR"})]}),"\n",(0,n.jsxs)(i.li,{children:["Added new Vertex AI text embedding model ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9476",children:"PR"})]}),"\n"]}),"\n",(0,n.jsx)(i.h2,{id:"llm-translation",children:"LLM Translation"}),"\n",(0,n.jsxs)(i.ul,{children:["\n",(0,n.jsxs)(i.li,{children:["OpenAI Web Search Tool Call Support ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9465",children:"PR"})]}),"\n",(0,n.jsxs)(i.li,{children:["Vertex AI topLogprobs support ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9518",children:"PR"})]}),"\n",(0,n.jsxs)(i.li,{children:["Support for sending images and video to Vertex AI multimodal embedding ",(0,n.jsx)(i.a,{href:"https://docs.litellm.ai/docs/providers/vertex#multi-modal-embeddings",children:"Doc"})]}),"\n",(0,n.jsxs)(i.li,{children:["Support litellm.api_base for Vertex AI + Gemini across completion, embedding, image_generation ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9516",children:"PR"})]}),"\n",(0,n.jsxs)(i.li,{children:["Bug fix for returning ",(0,n.jsx)(i.code,{children:"response_cost"})," when using litellm python SDK with LiteLLM Proxy ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/commit/6fd18651d129d606182ff4b980e95768fc43ca3d",children:"PR"})]}),"\n",(0,n.jsxs)(i.li,{children:["Support for ",(0,n.jsx)(i.code,{children:"max_completion_tokens"})," on Mistral API ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9606",children:"PR"})]}),"\n",(0,n.jsxs)(i.li,{children:["Refactored Vertex AI passthrough routes - fixes unpredictable behaviour with auto-setting default_vertex_region on router model add ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9467",children:"PR"})]}),"\n"]}),"\n",(0,n.jsx)(i.h2,{id:"spend-tracking-improvements",children:"Spend Tracking Improvements"}),"\n",(0,n.jsxs)(i.ul,{children:["\n",(0,n.jsxs)(i.li,{children:["Log 'api_base' on spend logs ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9509",children:"PR"})]}),"\n",(0,n.jsxs)(i.li,{children:["Support for Gemini audio token cost tracking ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9535",children:"PR"})]}),"\n",(0,n.jsxs)(i.li,{children:["Fixed OpenAI audio input token cost tracking ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9535",children:"PR"})]}),"\n"]}),"\n",(0,n.jsx)(i.h2,{id:"ui",children:"UI"}),"\n",(0,n.jsx)(i.h3,{id:"model-management",children:"Model Management"}),"\n",(0,n.jsxs)(i.ul,{children:["\n",(0,n.jsxs)(i.li,{children:["Allowed team admins to add/update/delete models on UI ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9572",children:"PR"})]}),"\n",(0,n.jsxs)(i.li,{children:["Added render supports_web_search on model hub ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9469",children:"PR"})]}),"\n"]}),"\n",(0,n.jsx)(i.h3,{id:"request-logs",children:"Request Logs"}),"\n",(0,n.jsxs)(i.ul,{children:["\n",(0,n.jsxs)(i.li,{children:["Show API base and model ID on request logs ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9572",children:"PR"})]}),"\n",(0,n.jsxs)(i.li,{children:["Allow viewing keyinfo on request logs ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9568",children:"PR"})]}),"\n"]}),"\n",(0,n.jsx)(i.h3,{id:"usage-tab",children:"Usage Tab"}),"\n",(0,n.jsxs)(i.ul,{children:["\n",(0,n.jsxs)(i.li,{children:["Added Daily User Spend Aggregate view - allows UI Usage tab to work > 1m rows ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9538",children:"PR"})]}),"\n",(0,n.jsxs)(i.li,{children:['Connected UI to "LiteLLM_DailyUserSpend" spend table ',(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9603",children:"PR"})]}),"\n"]}),"\n",(0,n.jsx)(i.h2,{id:"logging-integrations",children:"Logging Integrations"}),"\n",(0,n.jsxs)(i.ul,{children:["\n",(0,n.jsxs)(i.li,{children:["Fixed StandardLoggingPayload for GCS Pub Sub Logging Integration ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9508",children:"PR"})]}
1),"\n",(0,n.jsxs)(i.li,{children:["Track ",(0,n.jsx)(i.code,{children:"litellm_model_name"})," on ",(0,n.jsx)(i.code,{children:"StandardLoggingPayload"})," ",(0,n.jsx)(i.a,{href:"https://docs.litellm.ai/docs/proxy/logging_spec#standardlogginghiddenparams",children:"Docs"})]}),"\n"]}),"\n",(0,n.jsx)(i.h2,{id:"performance--reliability-improvements",children:"Performance / Reliability Improvements"}),"\n",(0,n.jsxs)(i.ul,{children:["\n",(0,n.jsxs)(i.li,{children:["LiteLLM Redis semantic caching implementation ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9356",children:"PR"})]}),"\n",(0,n.jsxs)(i.li,{children:["Gracefully handle exceptions when DB is having an outage ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9533",children:"PR"})]}),"\n",(0,n.jsxs)(i.li,{children:["Allow Pods to startup + passing /health/readiness when allow_requests_on_db_unavailable: True and DB is down ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9569",children:"PR"})]}),"\n"]}),"\n",(0,n.jsx)(i.h2,{id:"general-improvements",children:"General Improvements"}),"\n",(0,n.jsxs)(i.ul,{children:["\n",(0,n.jsxs)(i.li,{children:["Support for exposing MCP tools on litellm proxy ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9426",children:"PR"})]}),"\n",(0,n.jsxs)(i.li,{children:["Support discovering Gemini, Anthropic, xAI models by calling their /v1/model endpoint ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9530",children:"PR"})]}),"\n",(0,n.jsxs)(i.li,{children:["Fixed route check for non-proxy admins on JWT auth ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9454",children:"PR"})]}),"\n",(0,n.jsxs)(i.li,{children:["Added baseline Prisma database migrations ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9565",children:"PR"})]}),"\n",(0,n.jsxs)(i.li,{children:["View all wildcard models on /model/info ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9572",children:"PR"})]}),"\n"]}),"\n",(0,n.jsx)(i.h2,{id:"security",children:"Security"}),"\n",(0,n.jsxs)(i.ul,{children:["\n",(0,n.jsxs)(i.li,{children:["Bumped next from 14.2.21 to 14.2.25 in UI dashboard ",(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/pull/9458",children:"PR"})]}),"\n"]}),"\n",(0,n.jsx)(i.h2,{id:"complete-git-diff",children:"Complete Git Diff"}),"\n",(0,n.jsx)(i.p,{children:(0,n.jsx)(i.a,{href:"https://github.com/BerriAI/litellm/compare/v1.63.14-stable.patch1...v1.65.0-stable",children:"Here's the complete git diff"})})]})}function p(e={}){let{wrapper:i}={...(0,s.R)(),...e.components};return i?(0,n.jsx)(i,{...e,children:(0,n.jsx)(c,{...e})}):c(e)}},29815(e,i,l){"use strict";l.d(i,{A:()=>n});var t=l(74848);function n({img:e,alt:i="",style:l,...s}){if("string"==typeof e||e&&"default"in e){let n="string"==typeof e?e:e.default;return(0,t.jsx)("img",{src:n,alt:i,loading:"lazy",decoding:"async",style:l,...s})}let{src:r}=e,a=r.images||[];return(0,t.jsx)("img",{src:r.src,srcSet:a.map(e=>`${e.path} ${e.width}w`).join(", "),sizes:"(max-width: 996px) 100vw, 900px",width:r.width,height:r.height,alt:i,loading:"lazy",decoding:"async",style:{maxWidth:"100%",height:"auto",...l},...s})}l(96540)},28453(e,i,l){"use strict";l.d(i,{R:()=>r,x:()=>a});var t=l(96540);let n={},s=t.createContext(n);function r(e){let i=t.useContext(s);return t.useMemo(function(){return"function"==typeof e?e(i):{...i,...e}},[i,e])}function a(e){let i;return i=e.disableParentContext?"function"==typeof e.components?e.components(n):e.components||n:r(e.components),t.createElement(s.Provider,{value:i},e.children)}}}]);
Line numbers count LF bytes from the start of the resource, as the search results do. Vendor segments are library code the classifier recognised; they are stored but not indexed. Bytes are shown as Latin1 characters, one per byte.