MCP Configuration

MCP Server

deepset_mcp.mcp.ToolConfig `dataclass`

Configuration for tool registration.

It allows users to define what arguments should be passed to the tool at registration time. These arguments will not be provided by the LLM as the tool will receive them programmatically through partial application.

The configuration also determines if a tool should store outputs in the object store.

needs_client `class-attribute` `instance-attribute`

needs_client: bool = False

If the tool should receive a configured instance of the 'AsyncDeepsetClient' at tool-registration time.

needs_workspace `class-attribute` `instance-attribute`

needs_workspace: bool = False

If the tool should receive a static deepset workspace at tool-registration time.

memory_type `class-attribute` `instance-attribute`

memory_type: MemoryType = NO_MEMORY

The type of memory this tool should use.

custom_args `class-attribute` `instance-attribute`

custom_args: dict[str, Any] = field(default_factory=dict)

Any other arguments that should be passed to the tool at registration time instead of being passed by the LLM.

deepset_mcp.mcp.configure_mcp_server

configure_mcp_server(
    mcp_server_instance: FastMCP,
    tools_to_register: set[str] | None = None,
    deepset_api_key: str | None = None,
    deepset_api_url: str | None = None,
    deepset_workspace: str | None = None,
    deepset_docs_shareable_prototype_url: str | None = None,
    get_api_key_from_authorization_header: bool = False,
    object_store_backend: str = "memory",
    object_store_redis_url: str | None = None,
    object_store_ttl: int = 600,
) -> None

Configure the MCP server with the specified tools and settings.

Parameters:

Name	Type	Description	Default
`mcp_server_instance`	`FastMCP`	The FastMCP server instance to configure	required
`tools_to_register`	`set[str] \| None`	Set of tool names to register with the server. Will register all tools if set to None.	`None`
`deepset_api_key`	`str \| None`	Optional Deepset API key for authentication	`None`
`deepset_api_url`	`str \| None`	Optional Deepset API base URL	`None`
`deepset_workspace`	`str \| None`	Pass a deepset workspace name if you only want to run the tools on a specific workspace.	`None`
`deepset_docs_shareable_prototype_url`	`str \| None`	Shareable prototype URL that allows access to a docs search pipeline. Will fall back to the default shareable prototype URL if set to None.	`None`
`get_api_key_from_authorization_header`	`bool`	Whether to extract API key from authorization header	`False`
`object_store_backend`	`str`	Object store backend type ('memory' or 'redis')	`'memory'`
`object_store_redis_url`	`str \| None`	Redis connection URL (required if backend='redis')	`None`
`object_store_ttl`	`int`	TTL in seconds for stored objects	`600`

Raises:

Type	Description
`ValueError`	If required parameters are missing or invalid

deepset_mcp.mcp.initialize_or_get_initialized_store `cached`

initialize_or_get_initialized_store(
    backend: str = "memory",
    redis_url: str | None = None,
    ttl: int = 600,
) -> ObjectStore

Initializes the object store or gets an existing object store instance if it was initialized before.

Parameters:

Name	Type	Description	Default
`backend`	`str`	Backend type ('memory' or 'redis')	`'memory'`
`redis_url`	`str \| None`	Redis connection URL (required if backend='redis')	`None`
`ttl`	`int`	Time-to-live in seconds for stored objects	`600`

Raises:

Type	Description
`ValueError`	If Redis backend selected but no URL provided
`Exception`	If Redis connection fails

deepset_mcp.mcp.build_tool

build_tool(
    base_func: Callable[..., Any],
    config: ToolConfig,
    api_key: str | None = None,
    workspace: str | None = None,
    use_request_context: bool = True,
    base_url: str | None = None,
    object_store: ObjectStore | None = None,
) -> Callable[..., Awaitable[Any]]

Universal tool creator that handles client injection, workspace, and decorators.

This function takes a base tool function and enhances it based on the tool's configuration.

Parameters:

Name	Type	Description	Default
`base_func`	`Callable[..., Any]`	The base tool function.	required
`config`	`ToolConfig`	Tool configuration specifying dependencies and custom arguments.	required
`api_key`	`str \| None`	The deepset API key to use.	`None`
`workspace`	`str \| None`	The workspace to use when using a static workspace.	`None`
`use_request_context`	`bool`	Whether to collect the API key from the request context.	`True`
`base_url`	`str \| None`	Base URL for the deepset API.	`None`
`object_store`	`ObjectStore \| None`	The ObjectStore instance to use for memory decorators.	`None`

Returns:

Type	Description
`Callable[..., Awaitable[Any]]`	An enhanced, awaitable tool function with an updated signature and docstring.

Object Store

deepset_mcp.tokonomics.RichExplorer

Presents Python objects in various Rich formats with navigation support.

Parameters:

Name	Type	Description	Default
`store`	`ObjectStore`	Object store used for lookups.	required
`max_items`	`int`	Maximum number of items to show for lists and nested collections (default: 20). Note: All keys are shown for top-level dicts.	`25`
`max_string_length`	`int`	Maximum string length before truncation (default: 300).	`300`
`max_depth`	`int`	Maximum depth for object representation (default: 3).	`4`
`max_search_matches`	`int`	Maximum number of search matches to display (default: 10).	`10`
`search_context_length`	`int`	Number of characters to show around search matches (default: 150).	`150`

init

__init__(
    store: ObjectStore,
    max_items: int = 25,
    max_string_length: int = 300,
    max_depth: int = 4,
    max_search_matches: int = 10,
    search_context_length: int = 150,
) -> None

Initialize the RichExplorer with storage and configuration options.

parse_reference

parse_reference(ref_str: str) -> tuple[str, str]

Parse @obj_id.path into (obj_id, path).

Parameters:

Name	Type	Description	Default
`ref_str`	`str`	Reference string like @obj_id.path or obj_id	required

Returns:

Type	Description
`tuple[str, str]`	Tuple of (obj_id, path)

explore

explore(obj_id: str, path: str = '') -> str

Return a string preview of the requested object.

Parameters:

Name	Type	Description	Default
`obj_id`	`str`	Identifier obtained from the store.	required
`path`	`str`	Navigation path using `.` or `[...]` notation (e.g. `@obj_id.path.to.attribute`).	`''`

Returns:

Type	Description
`str`	String representation of the object.

search

search(
    obj_id: str,
    pattern: str,
    path: str = "",
    case_sensitive: bool = False,
) -> str

Search for a pattern within a string object.

Parameters:

Name	Type	Description	Default
`obj_id`	`str`	Identifier obtained from the store.	required
`pattern`	`str`	Regular expression pattern to search for.	required
`path`	`str`	Navigation path to search within (optional).	`''`
`case_sensitive`	`bool`	Whether search should be case sensitive.	`False`

Returns:

Type	Description
`str`	Search results as formatted string.

slice

slice(
    obj_id: str,
    start: int = 0,
    end: int | None = None,
    path: str = "",
) -> str

Extract a slice from a string or list object.

Parameters:

Name	Type	Description	Default
`obj_id`	`str`	Identifier of the object.	required
`start`	`int`	Start index for slicing.	`0`
`end`	`int \| None`	End index for slicing (None for end of sequence).	`None`
`path`	`str`	Navigation path to object to slice (optional).	`''`

Returns:

Type	Description
`str`	String representation of the slice.

deepset_mcp.tokonomics.InMemoryBackend

In-memory backend with counter-based IDs.

init

__init__() -> None

Initialize an instance of InMemoryBackend.

generate_id

generate_id() -> str

Generate sequential ID.

set

set(
    key: str, value: bytes, ttl_seconds: int | None
) -> None

Set a value at key.

get

get(key: str) -> bytes | None

Get a value at key.

delete

delete(key: str) -> bool

Delete a value at key.

deepset_mcp.tokonomics.ObjectStore

JSON-based object store with pluggable backends.

init

__init__(
    backend: ObjectStoreBackend, ttl: int = 600
) -> None

Initialize ObjectStore with backend and TTL.

Parameters

backend : Backend implementation for storage ttl : Time-to-live in seconds for stored objects. The TTL is managed by the backend.

put

put(obj: Any) -> str

Store any object as JSON using backend-generated ID.

get

get(obj_id: str) -> Any | None

Get object as JSON-decoded data.

delete

delete(obj_id: str) -> bool

Delete object.

deepset_mcp.tokonomics.ObjectStoreBackend

Bases: Protocol

Backend protocol with ID generation.

generate_id

generate_id() -> str

Generate a unique ID for this backend.

set

set(
    key: str, value: bytes, ttl_seconds: int | None
) -> None

Store bytes value with optional TTL.

get

get(key: str) -> bytes | None

Retrieve bytes value or None if not found/expired.

delete

delete(key: str) -> bool

Delete and return True if existed.

deepset_mcp.tokonomics.RedisBackend

Redis backend with UUID-based IDs.

init

__init__(redis_url: str) -> None

Initialize the redis backend.

generate_id

generate_id() -> str

Generate UUID.

set

set(
    key: str, value: bytes, ttl_seconds: int | None
) -> None

Set a value at key.

get

get(key: str) -> bytes | None

Get a value at key.

delete

delete(key: str) -> bool

Delete a value at key.

deepset_mcp.tokonomics.explorable

explorable(
    *, object_store: ObjectStore, explorer: RichExplorer
) -> Callable[[F], F]

Decorator factory that stores function results for later reference.

Parameters:

Name	Type	Description	Default
`object_store`	`ObjectStore`	The object store instance to use for storage.	required
`explorer`	`RichExplorer`	The RichExplorer instance to use for previews.	required

Returns:

Type	Description
`Callable[[F], F]`	Decorator function. Examples -------- >>> store = ObjectStore() >>> explorer = RichExplorer(store) >>> >>> @explorable(object_store=store, explorer=explorer) ... def process_data(data: dict) -> dict: ... return {"processed": data} ... >>> result = process_data({"input": "value"}) >>> # result contains a preview and object ID like "@obj_123"

deepset_mcp.tokonomics.explorable_and_referenceable

explorable_and_referenceable(
    *, object_store: ObjectStore, explorer: RichExplorer
) -> Callable[[F], F]

Decorator factory that combines @explorable and @referenceable functionality.

The decorated function can accept reference parameters AND stores its result in the object store for later reference.

Parameters:

Name	Type	Description	Default
`object_store`	`ObjectStore`	The object store instance to use.	required
`explorer`	`RichExplorer`	The RichExplorer instance to use.	required

Returns:

Type Description

Callable[[F], F]

Decorator function. Examples -------- >>> store = ObjectStore() >>> explorer = RichExplorer(store) >>> >>> @explorable_and_referenceable(object_store=store, explorer=explorer) ... def merge_data(data1: dict, data2: dict) -> dict: ... return {data1, data2} ... >>> # Accepts references and returns preview with object ID >>> result = merge_data("@obj_123", {"new": "data"}) >>> # result contains a preview and can be referenced as "@obj_002"

deepset_mcp.tokonomics.referenceable

referenceable(
    *, object_store: ObjectStore, explorer: RichExplorer
) -> Callable[[F], F]

Decorator factory that enables parameters to accept object references.

Parameters can accept reference strings like '@obj_id' or '@obj_id.path.to.value' which are automatically resolved before calling the function.

Parameters:

Name	Type	Description	Default
`object_store`	`ObjectStore`	The object store instance to use for lookups.	required
`explorer`	`RichExplorer`	The RichExplorer instance to use for path validation.	required

Returns:

Type Description

Callable[[F], F]

Decorator function. Examples -------- >>> store = ObjectStore() >>> explorer = RichExplorer(store) >>> >>> @referenceable(object_store=store, explorer=explorer) ... def process_data(data: dict, threshold: int) -> str: ... return f"Processed {len(data)} items with threshold {threshold}" ... >>> # Call with actual values >>> process_data({"a": 1, "b": 2}, 10) >>> >>> # Call with references >>> process_data("@obj_123", "@obj_456.config.threshold")

MCP Configuration

MCP Server

deepset_mcp.mcp.ToolConfig dataclass

needs_client class-attribute instance-attribute

needs_workspace class-attribute instance-attribute

memory_type class-attribute instance-attribute

custom_args class-attribute instance-attribute

deepset_mcp.mcp.configure_mcp_server

deepset_mcp.mcp.initialize_or_get_initialized_store cached

deepset_mcp.mcp.build_tool

Object Store

deepset_mcp.tokonomics.RichExplorer

__init__

parse_reference

explore

search

slice

deepset_mcp.tokonomics.InMemoryBackend

__init__

generate_id

set

get

delete

deepset_mcp.tokonomics.ObjectStore

__init__

Parameters

put

get

delete

deepset_mcp.tokonomics.ObjectStoreBackend

generate_id

set

get

delete

deepset_mcp.tokonomics.RedisBackend

__init__

generate_id

set

get

delete

deepset_mcp.tokonomics.explorable

deepset_mcp.tokonomics.explorable_and_referenceable

deepset_mcp.tokonomics.referenceable

deepset_mcp.mcp.ToolConfig `dataclass`

needs_client `class-attribute` `instance-attribute`

needs_workspace `class-attribute` `instance-attribute`

memory_type `class-attribute` `instance-attribute`

custom_args `class-attribute` `instance-attribute`

deepset_mcp.mcp.initialize_or_get_initialized_store `cached`

init

init

init

init