{"licence":{"name":"CC BY-SA 4.0","spdx":"CC-BY-SA-4.0","url":"https://creativecommons.org/licenses/by-sa/4.0/","attribution":"Atlas, a bilingual technical dictionary (https://cmaintz.github.io/tech-atlas/)"},"id":"ai/model-context-protocol","url":{"en":"https://cmaintz.github.io/tech-atlas/en/terms/ai/model-context-protocol/","da":"https://cmaintz.github.io/tech-atlas/da/terms/ai/model-context-protocol/"},"term":{"en":"Model Context Protocol (MCP)","da":"Model Context Protocol (MCP)"},"aka":{"en":["MCP"],"da":["MCP"]},"domain":["ai"],"cluster":"agents","layer":"agent","status":"emerging","era":2024,"summary":{"en":"An open standard for plugging tools and data into AI assistants, so each tool is hooked up once and works in many apps.","da":"En åben standard for at koble værktøjer og data til AI-assistenter, så hvert værktøj kobles på én gang og virker i mange apps."},"body":{"formal":{"en":"An open protocol in which an AI app, acting as a client, connects to MCP servers that offer tools, readable data and ready-made prompts as JSON messages; the app lists what each server offers and passes tool calls to it. Begun by one AI company in 2024, it has been run by a neutral foundation under the Linux Foundation since December 2025.","da":"En åben protokol, hvor en AI-app som klient forbinder sig til MCP-servere, der tilbyder værktøjer, data til læsning og færdige prompts som JSON-beskeder; appen viser, hvad hver server tilbyder, og sender værktøjskald videre til den. Den blev startet af en AI-virksomhed i 2024 og drives siden december 2025 af en neutral fond under Linux Foundation."},"plain":{"en":"Like the standard wall plug - once every lamp and kettle uses the same shape, you can move any of them to any room without rewiring.","da":"Som den almindelige stikkontakt - når alle lamper og køkkenapparater bruger samme form, kan man flytte dem til ethvert rum uden at trække nye ledninger."},"inPractice":{"en":"The IT lead at a Danish upper-secondary school connects the support case system and the shared drive to an AI assistant by listing two MCP servers in a settings file; the same two servers then work unchanged in the staff chat app.","da":"Den IT-ansvarlige på et dansk gymnasium kobler skolens sagssystem til supporthenvendelser og det fælles drev til en AI-assistent ved at skrive to MCP-servere i en indstillingsfil; de samme to servere virker så uændret i personalets chatapp."},"whyItMatters":{"en":"It ends the need to build a separate link for every tool and every AI app, but it also makes it very easy to hand an assistant wide reach you have not checked.","da":"Det fjerner behovet for at bygge en særskilt forbindelse for hvert værktøj og hver AI-app, men gør det også meget let at give en assistent stor rækkevidde, man ikke har tjekket."}},"deepDive":{"en":"Anthropic published MCP in November 2024 as an answer to the M×N integration problem: M AI applications each needing bespoke connectors to N tools. Its design borrows openly from the Language Server Protocol, which solved the same problem for editors and language tooling. Three roles are defined. The host is the user-facing application (an IDE, chat client or agent runtime); it instantiates one client per server connection; each server exposes capabilities. All messages are JSON-RPC 2.0 requests, results, errors and notifications, carried over stdio for local servers or Streamable HTTP for remote ones. Servers offer tools, resources and prompts; clients have historically offered roots, sampling (letting a server ask the host's model for a completion) and elicitation (letting a server ask the user for input).\n\nThe specification is versioned by date, and each revision has changed security-relevant behaviour. 2024-11-05 was the initial release. 2025-03-26 added an OAuth 2.1-based authorization framework, replaced the original HTTP+SSE transport with Streamable HTTP, and introduced tool annotations. 2025-06-18 removed JSON-RPC batching, added structured tool output and elicitation, classified MCP servers as OAuth resource servers with protected-resource metadata, and required clients to send RFC 8707 resource indicators so that a token minted for one server cannot be replayed at another. 2025-11-25 added OAuth Client ID Metadata Documents, experimental tasks for long-running requests, URL-mode elicitation and tool use inside sampling. The 2026-07-28 revision, the largest so far, made the core stateless: no initialize handshake or protocol sessions, protocol version and client capabilities sent in every request's _meta, a mandatory server/discover method, a multi-round-trip pattern that replaces server-initiated requests, tasks moved into an official extension, and a formal lifecycle policy with at least twelve months' notice before removal. Roots, sampling and logging were deprecated in the same revision, and Dynamic Client Registration was deprecated in favour of Client ID Metadata Documents.\n\nGovernance moved on 9 December 2025, when Anthropic donated MCP to the newly formed Agentic AI Foundation under the Linux Foundation, co-founded with Block and OpenAI; changes now go through a public Specification Enhancement Proposal (SEP) process with working groups and tiered SDKs. At the time of the donation Anthropic reported more than 10,000 active public MCP servers.\n\nMCP standardises the interface, not the trust decision. The protocol cannot stop a host from exposing a malicious server's tool descriptions to the model, and the specification's consent and tool-safety principles are largely obligations on host implementers. Compared with vendor-specific function-calling APIs, MCP moves tool definitions out of the application and into independently deployed servers; compared with the Agent2Agent protocol, it connects an agent to capabilities rather than to peer agents with their own reasoning.","da":"Anthropic offentliggjorde MCP i november 2024 som svar på M×N-integrationsproblemet: M AI-applikationer, der hver især har brug for skræddersyede connectorer til N værktøjer. Designet låner åbent fra Language Server Protocol, der løste samme problem for editorer og sprogværktøjer. Der er defineret tre roller. Værten (host) er den brugervendte applikation (en IDE, en chatklient eller et agentmiljø); den opretter én klient pr. serverforbindelse; og hver server udstiller kapabiliteter. Alle beskeder er JSON-RPC 2.0-requests, -resultater, -fejl og -notifikationer, transporteret over stdio for lokale servere eller Streamable HTTP for fjernservere. Servere tilbyder tools, resources og prompts; klienter har historisk tilbudt roots, sampling (hvor en server kan bede værtens model om et svar) og elicitation (hvor en server kan bede brugeren om input).\n\nSpecifikationen versioneres efter dato, og hver revision har ændret sikkerhedsrelevant adfærd. 2024-11-05 var den første udgave. 2025-03-26 tilføjede et autorisationsrammeværk baseret på OAuth 2.1, erstattede den oprindelige HTTP+SSE-transport med Streamable HTTP og indførte tool annotations. 2025-06-18 fjernede JSON-RPC-batching, tilføjede struktureret værktøjsoutput og elicitation, klassificerede MCP-servere som OAuth-ressourceservere med protected resource metadata og krævede, at klienter sender resource indicators efter RFC 8707, så et token udstedt til én server ikke kan genbruges mod en anden. 2025-11-25 tilføjede OAuth Client ID Metadata Documents, eksperimentelle tasks til langvarige forespørgsler, elicitation i URL-tilstand og værktøjsbrug inde i sampling. Revisionen 2026-07-28, den hidtil største, gjorde kernen tilstandsløs: intet initialize-håndtryk eller protokolsessioner, protokolversion og klientkapabiliteter sendes i hver forespørgsels _meta, en obligatorisk server/discover-metode, et mønster med flere rundture, der erstatter forespørgsler startet af serveren, tasks flyttet til en officiel udvidelse og en formel livscykluspolitik med mindst tolv måneders varsel før fjernelse. Roots, sampling og logging blev udfaset i samme revision, og Dynamic Client Registration blev udfaset til fordel for Client ID Metadata Documents.\n\nStyringen skiftede den 9. december 2025, da Anthropic overdrog MCP til den nystiftede Agentic AI Foundation under Linux Foundation, grundlagt sammen med Block og OpenAI; ændringer går nu gennem en offentlig proces med Specification Enhancement Proposals (SEP'er), arbejdsgrupper og SDK'er inddelt i niveauer. Ved overdragelsen oplyste Anthropic, at der fandtes over 10.000 aktive offentlige MCP-servere.\n\nMCP standardiserer grænsefladen, ikke tillidsbeslutningen. Protokollen kan ikke forhindre en vært i at vise en ondsindet servers værktøjsbeskrivelser til modellen, og specifikationens principper om samtykke og værktøjssikkerhed er i vid udstrækning forpligtelser for dem, der implementerer værter. Sammenlignet med leverandørspecifikke API'er til function calling flytter MCP værktøjsdefinitionerne ud af applikationen og over i selvstændigt udrullede servere; sammenlignet med Agent2Agent-protokollen forbinder den en agent med kapabiliteter frem for med ligestillede agenter, der selv ræsonnerer."},"edges":[{"type":"requires","to":"cs/api","confidence":"high","strength":"normal"},{"type":"requires","to":"cs/client","confidence":"high","strength":"normal"},{"type":"kind-of","to":"cs/protocol","confidence":"high","strength":"normal"},{"type":"implements","to":"ai/tool-calling","why":{"en":"MCP turns tool calling into a shared, open standard, so a tool is described once and any app that speaks MCP can call it.","da":"MCP gør værktøjskald til en fælles, åben standard, så et værktøj beskrives én gang, og enhver app, der taler MCP, kan kalde det."},"confidence":"high","strength":"primary"},{"type":"used-with","to":"ai/retrieval-augmented-generation","why":{"en":"MCP servers are a common way to fetch documents from company systems into the model's view at answer time.","da":"MCP-servere er en udbredt måde at hente dokumenter fra virksomhedens systemer ind i modellens synsfelt, når den svarer."},"confidence":"medium","strength":"normal"}],"depth":5,"sources":[{"title":"Model Context Protocol specification","tier":"official-doc","publisher":"Model Context Protocol project (Agentic AI Foundation, Linux Foundation)"},{"title":"Anthropic (2024), Introducing the Model Context Protocol","tier":"reference","publisher":"Anthropic"},{"title":"Anthropic (2025), Donating the Model Context Protocol and establishing the Agentic AI Foundation","url":"https://anthropic.com/news/donating-the-model-context-protocol-and-establishing-of-the-agentic-ai-foundation","tier":"reference","publisher":"Anthropic"},{"title":"Model Context Protocol specification 2026-07-28 - Key Changes","url":"https://modelcontextprotocol.io/specification/2026-07-28/changelog","tier":"official-doc","publisher":"Model Context Protocol project"}],"draft":true}