Glossary
The 68 terms Model Context Protocol in Practice defines, each in the book's own words, with the section that defines it.
From Edition 2026.11.14, the book's current edition. Where a definition depends on a package, it says the version this edition pins and runs its examples on.
A
- access token
-
An access token is a bearer credential, so whoever holds it can present it.
Defined in Bind each token to one server, chapter 8. In the book since Edition 2026.11.14. Permanent link to access token
- application-driven
-
The specification calls resources application-driven: hosts decide how to use them, whether by letting the user pick, by their own heuristics, or by asking the model
Defined in Put resources in the model's context, chapter 10. In the book since Edition 2026.11.14. Permanent link to application-driven
- argument completion
-
Argument completion lets the host ask the server for suggestions while the user types.
Defined in Complete arguments for the user, chapter 4. In the book since Edition 2026.11.14. Permanent link to argument completion
- authorization server
-
an authorization server issues the tokens, either beside the MCP server or apart from it
Defined in Authorize clients with OAuth, chapter 8. In the book since Edition 2026.11.14. Permanent link to authorization server
C
- cache scope also cacheScope
-
cacheScope, either "public", which a cache may share across callers, or "private", which it MUST NOT share across authorization contexts
Defined in Cache list results, chapter 2. In the book since Edition 2026.11.14. Permanent link to cache scope
- capability
-
A capability is something a client or server declares it supports, such as a server that offers tools or a client that can answer a request for input.
Defined in Separate the host, the client and the server, chapter 2. In the book since Edition 2026.11.14. Permanent link to capability
- client
-
A client is the host's connection to one server
Defined in Separate the host, the client and the server, chapter 2. In the book since Edition 2026.11.14. Permanent link to client
- Client ID Metadata Document
-
A Client ID Metadata Document makes the client ID a URL.
- Implemented by: mcp 2.3.0
Defined in Identify the client with a metadata document, chapter 8. In the book since Edition 2026.11.14. Permanent link to Client ID Metadata Document
- confused deputy
-
In the classic sense, a confused deputy is a program that holds authority of its own, takes a name from its caller, and uses its authority on that name where the caller had none.
Defined in Keep the server from acting as a confused deputy, chapter 9. In the book since Edition 2026.11.14. Permanent link to confused deputy
- consent
-
Consent is the host asking the user before a call that changes something
Defined in Shape tools around tasks, not endpoints, chapter 3. In the book since Edition 2026.11.14. Permanent link to consent
- core protocol
-
The core protocol is what every client and server must implement.
Defined in Negotiate an extension, chapter 7. In the book since Edition 2026.11.14. Permanent link to core protocol
D
- destructiveHint also destructive hint
-
a tool that changes something may destroy or overwrite data, rather than only add to it
- Part of: tool annotation
Defined in Mark what a tool changes with annotations, chapter 3. In the book since Edition 2026.11.14. Permanent link to destructiveHint
- DNS rebinding
-
With DNS rebinding, an attacker's domain can be made to resolve to 127.0.0.1 after the page loads, so the browser treats the local server as part of the attacker's site.
Defined in Reject requests from other origins, chapter 6. In the book since Edition 2026.11.14. Permanent link to DNS rebinding
- dual-era
-
a dual-era server, which serves both eras on the same endpoint
Defined in Serve old and new clients through the window, chapter 12. In the book since Edition 2026.11.14. Permanent link to dual-era
E
- earliest removal
-
An earliest removal is the date when a feature becomes eligible to go, not when it goes.
Defined in Inventory what the old server depends on, chapter 12. In the book since Edition 2026.11.14. Permanent link to earliest removal
- elicitation
-
elicitation: a structured question that a server sends and the client shows the user
Defined in Ask the user in the middle of a call, chapter 5. In the book since Edition 2026.11.14. Permanent link to elicitation
- era
-
A client still learns its server's era: the pinned revision or a 2025 one.
Defined in Connect one client per server, chapter 10. In the book since Edition 2026.11.14. Permanent link to era
- extension
-
Tasks is an extension, an optional addition to the protocol that both sides must agree to use.
- Implemented by: mcp 2.3.0
Defined in Run long work as tasks, chapter 7. In the book since Edition 2026.11.14. Permanent link to extension
- extension identifier
-
An extension's identifier is a vendor prefix, a slash and a name.
Defined in Negotiate an extension, chapter 7. In the book since Edition 2026.11.14. Permanent link to extension identifier
F
- form mode
-
In form mode, the server sends a message and a schema, the client renders a form, and the answer returns through the client.
- A kind of: elicitation
Defined in Ask for structured input with elicitation, chapter 5. In the book since Edition 2026.11.14. Permanent link to form mode
H
- handle also handles, explicit handle
-
a tool that needs state from an earlier call gets it as an explicit handle: one tool returns an ID, and the next takes it as an argument
Defined in Shape tools around tasks, not endpoints, chapter 3. In the book since Edition 2026.11.14. Permanent link to handle
- host
-
The host is the application the user works in.
Defined in Separate the host, the client and the server, chapter 2. In the book since Edition 2026.11.14. Permanent link to host
- host loop
-
The book's host loop, lib.host, is a small host the book writes itself, so each step is visible.
Defined in Let a model choose the tool, chapter 1. In the book since Edition 2026.11.14. Permanent link to host loop
I
- idempotentHint also idempotent hint
-
calling the tool again with the same arguments has no further effect
- Part of: tool annotation
Defined in Mark what a tool changes with annotations, chapter 3. In the book since Edition 2026.11.14. Permanent link to idempotentHint
- implicit poisoning
-
In implicit poisoning, one tool's metadata steers the agent to a different, more privileged tool, and the poisoned tool is never called.
Defined in Catch poisoned tool descriptions, chapter 9. In the book since Edition 2026.11.14. Permanent link to implicit poisoning
- indirect prompt injection
-
Indirect prompt injection is the general name for data that carries instructions: text the application retrieves, here a customer's ticket, contains instructions that the model follows, and those instructions can decide which tools it calls
Defined in Treat ticket text as untrusted data, chapter 9. In the book since Edition 2026.11.14. Permanent link to indirect prompt injection
- input request
-
The server answers the call with an input request instead of a result, and the client calls again with the answer.
Defined in Ask the user in the middle of a call, chapter 5. In the book since Edition 2026.11.14. Permanent link to input request
M
- MCP Apps
-
MCP Apps is an official extension that changes how a result is shown.
- A kind of: extension
Defined in Know what MCP Apps add, chapter 7. In the book since Edition 2026.11.14. Permanent link to MCP Apps
- mix-up attack also mix-up
-
In a mix-up, an attacker's authorization server gets the host to start a flow with it, then sends the user's browser on to an honest authorization server.
Defined in Validate the issuer of each response, chapter 8. In the book since Edition 2026.11.14. Permanent link to mix-up attack
- Model Context Protocol also MCP
-
The Model Context Protocol (MCP) is how an application that uses a model gets tools and data from programs it didn't write.
- Implemented by: mcp 2.3.0
Defined in MCP end to end, chapter 1. In the book since Edition 2026.11.14. Permanent link to Model Context Protocol
- Multi Round-Trip Request
-
a Multi Round-Trip Request, a pattern in which one call can end with a question instead of a result
- Implemented by: mcp 2.3.0
Defined in Ask the user in the middle of a call, chapter 5. In the book since Edition 2026.11.14. Permanent link to Multi Round-Trip Request
O
- openWorldHint also open world hint
-
the tool reaches entities outside a closed domain, such as the web
- Part of: tool annotation
Defined in Mark what a tool changes with annotations, chapter 3. In the book since Edition 2026.11.14. Permanent link to openWorldHint
P
- PKCE also Proof Key for Code Exchange
-
PKCE binds the code to the client that asked for it.
- Implemented by: mcp 2.3.0
Defined in Get a token with PKCE, chapter 8. In the book since Edition 2026.11.14. Permanent link to PKCE
- primitive also primitives
-
A server's three primitives are its tools, resources and prompts
Defined in Put each capability in the right primitive, chapter 2. In the book since Edition 2026.11.14. Permanent link to primitive
- progress also notifications/progress
-
Progress is MCP's answer to a long call: the server reports how far it has got while the call runs
Defined in Stream progress on a long call, chapter 6. In the book since Edition 2026.11.14. Permanent link to progress
- progressive discovery
-
progressive discovery, a small search tool that returns definitions on demand
Defined in What goes wrong in production, chapter 10. In the book since Edition 2026.11.14. Permanent link to progressive discovery
- prompt also prompts, MCP prompt
-
A prompt is a template of messages, with arguments, that the server defines and the user picks
Defined in Prompts are templates the user chooses, chapter 2. In the book since Edition 2026.11.14. Permanent link to prompt
- Protected Resource Metadata
-
Protected Resource Metadata (RFC 9728), a JSON document that names the resource and lists at least one authorization server
- Implemented by: mcp 2.3.0
Defined in Find the authorization server from the resource, chapter 8. In the book since Edition 2026.11.14. Permanent link to Protected Resource Metadata
- protocol error also protocol errors
-
A protocol error is a JSON-RPC error, for an unknown tool, a request that doesn't fit the tools/call schema, or a server failure.
- Contrasts with: tool execution error
Defined in Report failures as tool results, chapter 3. In the book since Edition 2026.11.14. Permanent link to protocol error
R
- readOnlyHint also read-only hint
-
the tool doesn't change its environment
- Part of: tool annotation
Defined in Mark what a tool changes with annotations, chapter 3. In the book since Edition 2026.11.14. Permanent link to readOnlyHint
- requestState
-
requestState is an opaque string the client must send back.
Defined in Return an input request instead of a result, chapter 5. In the book since Edition 2026.11.14. Permanent link to requestState
- resource also resources
-
A resource is data a server exposes under a URI, and the application decides when to read it and what to do with it
- Contrasts with: tool
Defined in Resources are data the application chooses, chapter 2. In the book since Edition 2026.11.14. Permanent link to resource
- resource annotation also resource annotations
-
A resource's annotations are three hints the host may use or ignore
Defined in Expose the policies as resources, chapter 4. In the book since Edition 2026.11.14. Permanent link to resource annotation
- resource template
-
A resource template names many resources with one URI template, written in the syntax of RFC 6570
Defined in Address records with resource templates, chapter 4. In the book since Edition 2026.11.14. Permanent link to resource template
- rug pull
-
a rug pull, where a server changes a definition after the user approved it
- A kind of: tool poisoning
Defined in Catch poisoned tool descriptions, chapter 9. In the book since Edition 2026.11.14. Permanent link to rug pull
S
- server
-
A server exposes capabilities, such as tools, and knows nothing else about the host.
Defined in Separate the host, the client and the server, chapter 2. In the book since Edition 2026.11.14. Permanent link to server
- shadowing
-
shadowing, where one server's description changes how the model uses another server's tools
- A kind of: tool poisoning
Defined in Catch poisoned tool descriptions, chapter 9. In the book since Edition 2026.11.14. Permanent link to shadowing
- span
-
an application creates spans, each a named, timed operation with attributes and a parent
Defined in Trace calls with OpenTelemetry, chapter 11. In the book since Edition 2026.11.14. Permanent link to span
- stdio also stdio transport
-
server.run() serves over stdio, where the client starts the server as a subprocess and the two exchange messages on standard input and output
True for mcp 2.3.0, as this edition pins it.
- Implemented by: mcp 2.3.0
Defined in Wrap one endpoint as a tool, chapter 1. In the book since Edition 2026.11.14. Permanent link to stdio
- step-up
-
Step-up is the 403 from Scope tools by what the token allows and the client's answer to it: on insufficient_scope, the client authorizes again with the union of the scopes it asked for before and the scopes in the challenge, retries, and limits how many times it tries
- Implemented by: mcp 2.3.0
Defined in Ask for more scope when a call needs it, chapter 8. In the book since Edition 2026.11.14. Permanent link to step-up
- sticky sessions also sticky routing
-
sticky sessions, which route every request carrying the same Mcp-Session-Id to the replica that held the session
Defined in Run replicas without sticky sessions, chapter 6. In the book since Edition 2026.11.14. Permanent link to sticky sessions
- Streamable HTTP also Streamable HTTP transport
-
For a server that many hosts share, the Model Context Protocol (MCP) defines a second transport, Streamable HTTP.
Defined in Serve over Streamable HTTP, chapter 6. In the book since Edition 2026.11.14. Permanent link to Streamable HTTP
- structured content also structuredContent
-
structuredContent: the result's data as JSON that a program can read without parsing text
Defined in Return structured content with an output schema, chapter 3. In the book since Edition 2026.11.14. Permanent link to structured content
T
- task
-
a task: the call's work, whose status the client asks for as often as the server suggests, and whose result waits on the server until the client fetches it
Defined in Run long work as tasks, chapter 7. In the book since Edition 2026.11.14. Permanent link to task
- task handle
-
A task's handle is a CreateTaskResult: a result whose resultType is "task" instead of "complete".
Defined in Start a task instead of waiting, chapter 7. In the book since Edition 2026.11.14. Permanent link to task handle
- task lifetime
-
A task's lifetime is its ttlMs, counted from createdAt, or null for no limit, and the server may change it.
Defined in Expire and cancel tasks, chapter 7. In the book since Edition 2026.11.14. Permanent link to task lifetime
- Tasks extension
-
With the Tasks extension, the server answers the call at once with a handle to a task
- A kind of: extension
Defined in Run long work as tasks, chapter 7. In the book since Edition 2026.11.14. Permanent link to Tasks extension
- terminal status
-
The completed, failed and cancelled statuses are terminal, and once a task reaches one, its state doesn't change
Defined in Poll a task and fetch its result, chapter 7. In the book since Edition 2026.11.14. Permanent link to terminal status
- token passthrough
-
a server either accepts a token that was issued to someone else or forwards the token it received to the API behind it
Defined in Never pass a client's token through, chapter 9. In the book since Edition 2026.11.14. Permanent link to token passthrough
- tool also tools
-
A tool is a named action with a description and an input schema, and the model decides whether to call it and with what
- Contrasts with: resource
Defined in Tools are actions the model chooses, chapter 2. In the book since Edition 2026.11.14. Permanent link to tool
- tool annotation also tool annotations, ToolAnnotations
-
A tool's annotations tell the host what a call will do before it's made
Defined in Mark what a tool changes with annotations, chapter 3. In the book since Edition 2026.11.14. Permanent link to tool annotation
- tool definition
-
A tool's definition is text from its server that the host passes to the model: its description, input schema and annotations.
Defined in Catch poisoned tool descriptions, chapter 9. In the book since Edition 2026.11.14. Permanent link to tool definition
- tool execution error also tool execution errors, execution error
-
A tool execution error is an ordinary result with isError set to true, for an API failure, an invalid input or a broken business rule.
- Contrasts with: protocol error
Defined in Report failures as tool results, chapter 3. In the book since Edition 2026.11.14. Permanent link to tool execution error
- tool poisoning
-
Tool poisoning puts instructions for the model into a tool's definition.
Defined in Catch poisoned tool descriptions, chapter 9. In the book since Edition 2026.11.14. Permanent link to tool poisoning
- tool-selection accuracy also top-1 accuracy
-
tool-selection accuracy: the share of cases in which the model's first call uses the tool a correct agent would use, often called top-1 accuracy
Defined in Measure how often the right tool is chosen, chapter 11. In the book since Edition 2026.11.14. Permanent link to tool-selection accuracy
- trust boundary
-
A trust boundary is a place where data or authority passes from one party to another that shouldn't take it on faith.
Defined in Draw the trust boundaries, chapter 9. In the book since Edition 2026.11.14. Permanent link to trust boundary
- TTL also ttlMs
-
ttlMs, how many milliseconds the result stays fresh, like HTTP's max-age
Defined in Cache list results, chapter 2. In the book since Edition 2026.11.14. Permanent link to TTL
U
- URL mode
-
In URL mode, the server sends the user to a web page, not a form
- A kind of: elicitation
Defined in Ask for structured input with elicitation, chapter 5. In the book since Edition 2026.11.14. Permanent link to URL mode