* Add FastMCPApp — a Provider for composable MCP applications
* Wire Prefab callable resolver via to_json(tool_resolver=) parameter
* Remove inspect.signature compat check, use try/except until prefab 0.10.0
* Address review: fix add_tool registry gaps, normalize auth errors, bump prefab to 0.10.0
* Register global key after _add_component succeeds
* Simplify: extract decorator dispatch, use get_fastmcp_meta, expose get_global_tool
* Remove prek from Marvin workflows
These workflows run Claude to respond to /marvin mentions — linting
the repo is unnecessary and fails without renderer deps installed.
* Return ResolvedTool from callable resolver, add contacts example
The callable resolver now returns ResolvedTool (from prefab_ui) instead of a
plain string, carrying metadata like unwrap_result that the renderer needs to
correctly handle structuredContent envelopes. The unwrap_result flag is derived
from the tool's x-fastmcp-wrap-result output schema marker.
* Bump prefab-ui requirement to >=0.11.0
* Remove stale ty ignore comments now that prefab-ui 0.11 is published
* Add tests for two-stage pattern, empty full-detail results, empty inputs
* Add ListTools, search limit, catalog size annotation; split tests
Co-authored-by: Claude <noreply@anthropic.com>
* Remove BM25 internal cap so Search.limit is the sole truncation point
* Pass default_limit to BM25 instead of arbitrary high cap
---------
Co-authored-by: Claude <noreply@anthropic.com>
Adds the PropelAuthProvider which delegates to the IntrospectionTokenVerifier
and optionally does an additional resource check.
Adds an example server and client which makes an authenticated request
and gets information from the token.
Updates the documentation (but only for v3 as this isn't in v2).
* Add prefab auto-wiring for MCP Apps (#3119)
Tools that return prefab types (UIResponse, Component) automatically get
wired to the shared prefab renderer resource. Works via app=True,
return type inference, or both.
* Prefab compatibility updates
* Use published prefab-ui >=0.6.0, remove local source override
* Migrate UIResponse to PrefabApp for Prefab UI integration
PrefabApp is a pure data object with to_json(), html(), and csp()
methods. Tools can return PrefabApp, bare Components, or
ToolResult with structured_content for custom LLM fallback text.
* Add Prefab UI apps documentation
* Add mini apps and full apps documentation pages
Mini apps covers the common single-screen patterns: charts (bar, line,
area, pie), data tables with sorting/search/pagination, forms (manual
and Pydantic-generated), status displays, conditional content, and
layout composition with tabs and accordions.
Full apps covers multi-page applications using Pages/Page components,
shared state across pages, and using ToolCall with result_key for
server-driven state updates.
* Reframe apps docs around motivation, add generative UIs page
The docs now lead with the problem — MCP tools stuff data into the LLM
context window, and building HTML/JS/CSS frontends is a non-starter for
Python developers — before introducing Prefab as the solution. Mini apps
are framed as the primary use case: focused, single-purpose UIs that
present data visually and collect structured input.
New generative UIs page covers the concept of LLMs producing component
JSON directly, enabling adaptive dashboards, tailored forms, and
exploratory workflows.
* Tag Prefab docs pages as SOON instead of NEW
* Rename Low-Level API to Custom HTML Apps
The page is about using the MCP Apps extension directly, not a FastMCP
or Prefab internal API. Reframed to make clear this is the open MCP
protocol with FastMCP providing convenience wrappers.
* Tighten apps docs and widen content area
Strip editorial motivation from all app doc pages — let code examples
do the talking. Add content-area max-width override (44rem) to style.css.
* Restructure apps docs, fix code issues
Rename Prefab UI → Prefab Apps, mini-apps → patterns, remove
generative-uis and full-apps pages. Rewrite prefab page to lead with
what users do (declare a UI, return it) before explaining internals.
Patterns page now has fully self-contained copy-pasteable examples with
explicit imports and links to prefab docs. Forms show the two-tool
pattern (form + handler). Add patterns_server.py example.
Code fixes: move get_args to module-level import, remove dead
AuthCheckCallable type alias, fix ToolCall→CallTool in all docs.
* Remove unused ToolResult import from chart_server
* Handle composite Prefab types in type inference and schema suppression
_has_prefab_return_type and the output schema suppression logic only
checked bare classes, missing unions (Column | None) and Annotated
wrappers (Annotated[PrefabApp | None, ...]). Recurse through Union,
types.UnionType, and Annotated to detect Prefab types in composite
annotations.
* code mode
* update uv.lock for monty optional dep
🤖 Generated with Claude Code
* retry CI
* Address PR review comments on CodeMode transform
🤖 Generated with Claude Code
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Fix ty unresolved-attribute error on search_helper
🤖 Generated with Claude Code
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* more idiomacy
* harden
* fix docs
* harden
* fix red CI
* Refactor CodeMode to use CatalogTransform base class
Removes the duplicate ContextVar bypass pattern in favor of the shared
CatalogTransform machinery. Also fixes a pre-existing bug where
`from __future__ import annotations` caused NameError for Annotated
in nested function scopes at runtime.
* Remove redundant _get_visible_tools wrapper in CodeMode
* Rewrite CodeMode docs with proper motivation and structure
* Fix type narrowing in collision test
* Stop unwrapping tool results in CodeMode's call_tool
call_tool() inside execute blocks now returns structured content as-is,
preserving the {"result": value} wrapping. This means the output schema
shown in search results accurately describes what call_tool() returns,
so LLMs can trust the schema when writing code.
Also adds examples/code_mode/ with a server and narrated client demo.
* Simplify call_tool return type: dict | str
* Fix example client to unwrap structured results
* Let server resolve tool versions instead of pinning first match
* Rewrite CodeMode docs to match current behavior
* Rename optional extra from monty to code-mode
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Jeremiah Lowin <153965+jlowin@users.noreply.github.com>
* feat: Add search transforms for tool discovery
RegexSearchTransform and BM25SearchTransform collapse large tool
catalogs into a search interface so LLMs discover tools on demand
instead of receiving the full listing.
* chore: Update SDK documentation
* fix: call_tool recursion guard, atomic BM25 rebuild, hash includes descriptions
* Extract CatalogTransform base class for catalog-aware transforms
Transforms that replace list_tools() with synthetic components (like
search) need to read the real catalog at call time without triggering
their own replacement logic. CatalogTransform handles the re-entrant
bypass via per-instance ContextVar, exposing transform_tools() as the
subclass hook and get_tool_catalog() for catalog access.
* Add search transform examples for regex and BM25
* Add README for search transform examples
* Polish search example clients with rich output
* Remove hardcoded tool counts from search example subtitles
* Clarify that review bot feedback should be evaluated on its merits
* Expand search transform docs with proper hierarchy
---------
Co-authored-by: marvin-context-protocol[bot] <225465937+marvin-context-protocol[bot]@users.noreply.github.com>
* Update repository references from jlowin/fastmcp to prefecthq/fastmcp
* Retrigger CI after repo transfer
* chore: Update SDK documentation
* Only run deep triage on bug issues for jlowin
---------
Co-authored-by: marvin-context-protocol[bot] <225465937+marvin-context-protocol[bot]@users.noreply.github.com>
When a background task calls ctx.elicit(), the notification subscriber now
detects the input_required notification and sends a standard elicitation/create
request to the client via session.elicit(). The client's elicitation_handler
fires, and the relay pushes the response to Redis for the blocked worker.
This means clients can respond to background task elicitation using the same
elicitation_handler they'd use for any other elicitation — no need to interact
with Redis or call handle_task_input() directly.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Rename ui= to app= and consolidate ToolUI/ResourceUI into AppConfig
* Remove backward-compat aliases for ToolUI/ResourceUI/ui_to_meta_dict
* Add extra=allow to AppConfig model_config for forward compatibility
- Add return type annotation to main() in run_with_tracing.py
- Use spread operator for argv construction
- Add type annotations to docs test example
- Use async httpx client and asyncio.sleep in diagnostics server
- Improve subprocess termination handling with timeout fallback
🤖 Generated with Claude Code
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Be specific about which operations are traced (tools, prompts, resources, resource templates)
- Remove "(not the SDK)" parenthetical
- Consolidate attribute documentation - remove redundancy in Tracing section
- Delete unnecessary examples/diagnostics/__init__.py
🤖 Generated with Claude Code
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Adds opt-in distributed tracing via OpenTelemetry for observability into
FastMCP server and client operations.
Server spans are created for tool calls, resource reads, and prompt
renders with attributes like component key, component type, provider
type, session ID, and auth context. Client spans wrap outgoing calls
with trace context propagation via W3C headers in request meta.
Components provide their own span attributes through a `get_span_attributes()`
method that subclasses override - this lets LocalProvider, FastMCPProvider,
and ProxyProvider each include relevant context (original names, backend URIs).
To enable: configure an OpenTelemetry SDK with a TracerProvider before
importing fastmcp. Traces export to any OTLP-compatible backend.
Closes ENG-2813
🤖 Generated with Claude Code
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* Add test_custom_subclass_tasks.py
* Refactor provider execution: delegate to middleware via wrapper components
- Remove execution methods (call_tool, read_resource, etc.) from Provider base
- Add FastMCPProvider* wrapper classes that delegate to child server middleware
- Move task routing to Tool._run() using contextvars (_task_metadata, _tool_call_key)
- Add convert_to_tool_result(result, output_schema) utility for Docket results
- Add convert_to_prompt_result() utility for prompt task results
- Pass namespaced key via add_to_docket(name=) for mounted tool lookup
* Standardize add_to_docket() with fn_key/task_key parameters
All components now use explicit fn_key (function lookup) and task_key
(result storage) parameters instead of relying on implicit key handling.
This fixes mounted component task execution where the MCP-visible key
differs from the Docket-registered function name.
* Add middleware chain tests for three-level mount hierarchy
Tests verify middleware runs at parent, child, and grandchild levels
for tools, resources, prompts, and resource templates.
* WIP: Provider refactor - unified submit_to_docket, template _read() in progress
Work in progress on refactoring execution to use component _read()/_run()/_render() methods.
Template background tasks not yet working - needs fix for Docket key lookup.
* Fix conversion functions to take full component for attribute access
Pass Tool/Prompt/Resource/Template to conversion functions instead of
individual attributes, ensuring access to serializer, output_schema,
mime_type, etc. Also fixes mixed-content output schema validation.
* Refactor: unified convert_result() methods and check_background_task helper
- Add convert_result() instance methods to all component types (Tool, Prompt, Resource, ResourceTemplate)
- Extract duplicated task routing logic into check_background_task() helper
- Fix type annotations on FastMCPProviderResource.read() and FastMCPProviderPrompt.render()
- Update protocol.py to use component.convert_result() uniformly
* Update tests to use namespace= instead of deprecated prefix= parameter
* Simplify Provider interface and consolidate docket registration
- Remove get_http_routes from Provider (unused)
- Remove ProviderLifespanConfig, _base_lifespan, _register_tasks
- Remove supports_tasks flag from Provider.__init__
- Consolidate all docket registration in server._docket_lifespan()
- Simplify lifespan() to take no parameters
- Move MountedProvider to separate module
* Fix control flow in ComponentService resource methods
* Move providers to server/providers
* Ensure MountedProvider get_* methods go through middleware
* Fix get_resource to only return concrete resources
Reverts template-checking in get_resource that broke task execution.
Tasks need access to the original template, not instantiated resources.
* Move prefix utilities into mounted.py, deprecate import_server
- Add resource prefix functions (add/remove/has_resource_prefix) to mounted.py
- Deprecate import_server with warning to use mount() instead
- Add tool_names uniqueness validation in MountedProvider
* Fix provider iteration order and remove dead _is_mounted flag
- Remove unused _is_mounted flag (MountedProvider.lifespan() calls _lifespan
not _lifespan_manager, so the flag was never checked)
- Fix provider iteration: change reversed() to forward order in execution
methods (_call_tool, _read_resource_middleware, _get_prompt_content_middleware)
to match documented "first non-None wins" semantics
- Fix ComponentService to handle prefix-less mounted servers using
_strip_tool_prefix()/_strip_resource_prefix() methods
- Update conflict resolution tests to expect first-registered provider wins
- Add regression tests for Docket behavior and prefix-less ComponentService
* Add TaskComponents type and exception handling for provider task registration
- Create TaskComponents dataclass with FunctionTool/FunctionResource/etc. types
for proper typing of get_tasks() return value
- Add try/except wrapper around provider.get_tasks() in _docket_lifespan for
consistent error handling (warn + continue or raise based on settings)
- Remove type: ignore comments from server.py task registration loop
* Fix tool_choice to always require tools when result_type is set
* Consolidate sampling examples with rich output
* Replace eval() with explicit add/multiply tools
* Unify SamplingHandler and promote OpenAI handler
Consolidates ServerSamplingHandler and ClientSamplingHandler into a single
SamplingHandler type alias. Moves OpenAISamplingHandler from experimental
to fastmcp.client.sampling.handlers.openai as the canonical location.
Backwards compatibility maintained for imports from experimental.
* Remove unreachable code paths in OpenAI handler
* Fix docstring and use elif for mutually exclusive branches
* MCP → SDK (vocab change only)
* WIP: Sampling API with SamplingResult[T] and result_type
* SEP-1577: Sampling with tools
- Add tools and result_type parameters to ctx.sample()
- Update OpenAI handler for tool content types
- Client advertises sampling.tools capability by default
- Collect tool results into single message with list content
* Fix tool result content handling in OpenAI handler
* Remove @sampling_tool decorator - pass functions directly to sample()
Functions passed to ctx.sample(tools=[...]) are now auto-converted
via SamplingTool.from_function(). Users can still use that method
directly for custom name/description overrides.
* Remove auto-conversion of MCP tools to sampling tools
Users want MCP tools passed to ctx.sample() to go through the full MCP
machinery (middleware, native responses) rather than being auto-converted
to direct function calls. Now only SamplingTool and plain callables are
accepted - passing a FastMCP Tool raises a clear TypeError.
Also bumps mcp dependency to >=1.24.0 for required sampling features.
* Refactor sampling API: replace sample_iter() with sample_step()
Replace the mutable SampleRun/sample_iter() pattern with a simpler stateless
sample_step() function. sample_step() makes a single LLM call and returns a
SampleStep with the response and history. sample() now loops sample_step()
internally.
Key changes:
- Add sample_step() for fine-grained control over the sampling loop
- Remove SampleRun class and sample_iter() method
- Structured output uses tool description only (no prompt modification)
- execute_tools parameter controls automatic vs manual tool execution
* Address CodeRabbit nitpicks
* Address CodeRabbit review feedback for sampling tools
- Fix temperature=0.0 being dropped due to falsy evaluation
- Add ToolChoice.name support for forcing specific tools
- Replace assert statements with explicit RuntimeError checks
- Add mask_error_details parameter to sample()/sample_step() with ToolError escape hatch
- Fix hasattr patterns with proper isinstance checks
- Document mask_error_details and add OpenAI prerequisites to docs
* Address additional CodeRabbit review feedback
- Catch ValidationError specifically instead of bare Exception
- Update result_type docs to mention dataclasses and basic types
- Raise ValueError for unknown tool_choice modes
- Validate sampling_handler_behavior to catch typos
- Remove ToolChoice.name handling (not part of MCP spec)
- Validate tool_choice string in sample_step()
* Review fixes for sampling tools PR
- Remove internal functions from sampling __init__.py exports
- Remove fragile is_text property, use not is_tool_use instead
- Inline call_client into context.py, remove from run.py
- Fix SamplingMessage docs to use TextContent
- Handle result.text being None in doc examples
- Simplify client sampling docs to recommend OpenAISamplingHandler
- Add sampling_capabilities override documentation
- Raise iteration limit from 50 to 100
- Remove _parse_model_preferences duplication
- Use AsyncOpenAI in OpenAISamplingHandler
- Fix tool_choice docstring
* Fix OpenAI handler tests to use AsyncOpenAI
* Address remaining CodeRabbit review comments
- Fix message ordering in OpenAI handler: tool results now correctly
follow assistant message with tool_calls
- sample_step() now always includes assistant message in history
- Raise ValueError on JSON parse errors instead of silent {}
- Add has_sampling capability check when behavior is None
- Raise RuntimeError when structured output receives text response
- Wrap primitive result_type schemas in object wrapper
- Fix docs example using invalid SamplingMessage construction
- Add comprehensive client_sampling_test.py example
* Add return type annotation to OpenAISamplingHandler.__init__
* Use explicit 'is not None' check for sampling_capabilities defaulting
---------
Co-authored-by: claude[bot] <41898282+claude[bot]@users.noreply.github.com>
Co-authored-by: Bill Easton <strawgate@users.noreply.github.com>
The task protocol (SEP-1686) is now always enabled - server always
registers task handlers and advertises task capabilities. Users still
opt into background execution at the server level (tasks=True) or
component level (task=True on tools, prompts, resources).
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Docket provides background task execution and is now always available
for all FastMCP servers. Only `enable_tasks` remains to control the
SEP-1686 task protocol support.
Changes:
- Remove `enable_docket` setting and related validation
- Docket/Worker lifecycle is always active in server lifespan
- CurrentDocket and CurrentWorker dependencies work without config
- Add server readiness signaling via `_started` event
- Fix test timing issues with proper port probing
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
* Implement MCP background tasks (SEP-1686) using Docket
Adds support for background task execution via the MCP task protocol,
powered by Docket for task queue management.
- Tools, resources, and prompts can be marked with `task=True` to run async
- Progress dependency for tracking task progress
- CurrentDocket and CurrentWorker dependencies for advanced use cases
- Client API with `.call_tool(..., task=True)` returns task handles
- Task status notifications via subscriptions
- CLI worker command for distributed task processing
Configuration via environment:
- FASTMCP_ENABLE_DOCKET=true
- FASTMCP_ENABLE_TASKS=true
- FASTMCP_DOCKET_URL=redis://... (or memory:// for single-process)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
* Fix tasks example import (TaskStatusResponse → GetTaskResult)
The example was using a non-existent TaskStatusResponse type.
Updated to use mcp.types.GetTaskResult which is what the
on_status_change callback actually receives.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
* Fix env var name in Docket error messages
The error messages referenced FASTMCP_EXPERIMENTAL_ENABLE_DOCKET but the
actual setting is FASTMCP_ENABLE_DOCKET.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
* Remove deprecated code re-added from pre-#2329 branch
- Remove ExtendedEnvSettingsSource (FASTMCP_SERVER_ prefix support)
- Remove dependencies parameter from FastMCP.__init__
* Replace fakeredis git pin with PyPI release
* Remove redundant fakeredis dev dep (pulled via pydocket)
---------
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Jeremiah Lowin <153965+jlowin@users.noreply.github.com>
* Fix RFC 8414 path-aware authorization server metadata discovery
Override get_well_known_routes() in OAuthProvider to rewrite the
authorization server metadata route to be path-aware based on issuer_url,
matching how protected resource metadata already works.
Closes#2527
* Update readme
