Skip to main content
Glama

flutter-mcp

MCP License: MIT Source: api.flutter.dev

Live Flutter, Dart SDK and pub.dev documentation for AI coding agents. The MCP server inside is published as flutter-docs-mcp; the repository is flutter-mcp.

An MCP (Model Context Protocol) server that gives AI coding agents live Flutter, Dart SDK and pub.dev documentation instead of whatever the model happens to remember. Pages are scraped from the official docs at request time, cached locally, and returned as clean markdown — so generated Flutter code targets real, current APIs rather than deprecated or invented ones.

  • Flutter classes — scraped live from api.flutter.dev (widgets, material, cupertino, foundation, services, rendering, …)

  • Dart SDK classes — scraped live from api.dart.dev (dart:core, dart:async, dart:collection, dart:io, …)

  • pub.dev packages — metadata + README from the pub.dev API

  • Local search index over ~4,500 Flutter / Dart SDK classes (46 libraries, dart:convert / dart:isolate / dart:ffi / dart:typed_data / dart:ui included), rebuilt at most every 7 days, with a stale fallback when the network fails

  • SQLite TTL cache so repeated lookups are instant

  • Politeness layer — robots.txt (RFC 9309), per-host throttle incl. Crawl-delay, Retry-After, conditional GET and request budgets (details)

The six tools

Tool

What it does

flutter_docs

Resolve one identifier (ListView, material.AppBar, dart:async.Future, pub:dio) to a single documentation page as markdown.

flutter_search

Ranked name search over the local index of all Flutter + Dart SDK classes, enums, mixins and typedefs.

flutter_mentions

Parse every @flutter_mcp … mention in a text and resolve each one — including version constraints like provider:^6.0.0 or dio:>=5.0.0 <6.0.0 — to exactly one docs entry.

pub_package

pub.dev package metadata (version, publisher, likes, pub points) plus its README as markdown.

flutter_status

Real health check: index size/age, cache stats, live probes of api.flutter.dev and pub.dev, and politeness counters.

health_check

Server liveness and version.

Every tool returns a plain dict. Failures come back as {"error": …, "suggestion": …} — a bad lookup never raises into the MCP layer, and no tool delegates to another tool.

Related MCP server: DevDocs MCP Server

Requirements

  • Python 3.10 or newer (developed and tested on 3.12)

  • uv for the one-command install below (uvx ships with it)

  • The MCP Python SDK 1.x is pinned (mcp>=1.2.0,<2.0); this package does not support the mcp 2.x rename of FastMCP

Install & run

One command — no checkout, no token

uvx --from git+https://github.com/KEEPEE/flutter-mcp.git flutter-docs

This builds the package in an isolated environment and starts the stdio MCP server. It prints nothing on purpose: stdout is the protocol channel. Stop it with Ctrl-C.

After the repository is updated, force uv to re-resolve the commit:

uvx --refresh --from git+https://github.com/KEEPEE/flutter-mcp.git flutter-docs

Local checkout — fastest startup, editable while developing

git clone https://github.com/KEEPEE/flutter-mcp.git
cd flutter-mcp
python3 -m venv .venv
.venv/bin/pip install -e .
.venv/bin/flutter-docs          # stdio MCP server

MCP client configuration

Generic stdio client (Claude Desktop, Cursor, Cline, …)

{
  "mcpServers": {
    "flutter-docs": {
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/KEEPEE/flutter-mcp.git",
        "flutter-docs"
      ]
    }
  }
}

With an explicit cache directory (any env you set is passed straight through to the server):

{
  "mcpServers": {
    "flutter-docs": {
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/KEEPEE/flutter-mcp.git",
        "flutter-docs"
      ],
      "env": {
        "FLUTTER_DOCS_MCP_CACHE_DIR": "/tmp/flutter-docs-cache"
      }
    }
  }
}

DeepSeek Harness (cordis-style plugin list)

- id: mcp-flutter-docs
  name: '@deepseek-ai/dsh-mcp-client'
  config:
    serverName: flutter-docs
    transport: stdio
    command: uvx
    args:
      [
        '--from',
        'git+https://github.com/KEEPEE/flutter-mcp.git',
        'flutter-docs'
      ]

To skip the build at client start, point command at the console script of a checkout instead — command: /path/to/flutter-mcp/.venv/bin/flutter-docs with args: [].

Tools reference

flutter_docs(identifier, topic=None, max_tokens=8000)

Resolves an identifier to one documentation page. Identifier forms, tried in this order:

Form

Meaning

pub:dio, pub:dio:5.11.1

pub.dev package: description + README

dart:async.Future

Dart SDK class on api.dart.dev (the library keeps its colon)

material.AppBar, widgets.ListView

Flutter class on api.flutter.dev, library given

ListView

exact match in the local index → widgets/material fallback → pub.dev

topic keeps only the section whose heading contains it (methods, properties, examples, constructors, …) plus the class title; if nothing matches, the full page is returned with a note listing the available headings. max_tokens is a rough budget (1 token ≈ 4 characters): the markdown is cut at a line boundary.

truncated is true whenever the returned content is not the whole page — a token-budget cut, a topic filter, or both — and truncation then says which, names the sections that were dropped, and gives tokens (returned), source_tokens (the whole page) and tokens_payload (the payload including its own [truncated: …] marker).

Returns {"type", "identifier", "url", "title", "content", "truncated", "tokens", "source_tokens", "max_tokens", "cached"}, plus truncation when something was cut and an optional note.

// flutter_docs({ "identifier": "material.AppBar", "topic": "properties", "max_tokens": 300 })
{
  "type": "flutter_class",
  "identifier": "material.AppBar",
  "url": "https://api.flutter.dev/flutter/material/AppBar-class.html",
  "title": "AppBar class",
  "content": "# AppBar class\n\nA Material Design app bar.\n\nAn app bar consists of a toolbar and potentially other widgets, such as a\n[TabBar](https://api.flutter.dev/flutter/material/TabBar-class.html) and a [FlexibleSpaceBar](https://… [truncated: showing ~273 of ~7367 estimated tokens]",
  "truncated": true,
  "tokens": 273,
  "source_tokens": 7367,
  "max_tokens": 300,
  "cached": true,
  "truncation": {
    "reasons": ["topic_filter", "token_budget"],
    "sections_dropped": ["Troubleshooting", "Constructors", "Methods", "Operators", "Static Methods"],
    "tokens_returned": 273,
    "tokens_source": 7367,
    "tokens_payload": 286,
    "max_tokens": 300,
    "explanation": "topic 'properties' dropped 6 section(s): …; token budget 300: body cut at a line boundary, ~273 of ~7367 estimated tokens returned"
  }
}
// flutter_docs({ "identifier": "dart:async.Future", "topic": "methods", "max_tokens": 250 })
{
  "type": "dart_class",
  "identifier": "dart:async.Future",
  "url": "https://api.dart.dev/dart-async/Future-class.html",
  "content": "# Future<T> class abstract interface\n\nThe result of an asynchronous computation. … ## Methods\n\nasStream() → Stream<T> …",
  "truncated": true,
  "cached": true,
  "title": "Future< T > class abstract interface"
}

A name that resolves nowhere reports every place it looked:

// flutter_docs({ "identifier": "NopeXYZ" })
{
  "error": "could not resolve 'NopeXYZ': flutter widgets: not found (HTTP 404): https://api.flutter.dev/flutter/widgets/NopeXYZ-class.html; flutter material: not found (HTTP 404): …; pub.dev: not found (HTTP 404): …",
  "suggestion": "try flutter_search('NopeXYZ') to find the right name"
}

flutter_search(query, limit=8)

Ranks the local index by name similarity (exact > prefix > substring > fuzzy). Returns {"query", "count", "results": [{name, library, kind, url, score}], "index_stale"}.

// flutter_search({ "query": "text field", "limit": 3 })
{
  "query": "text field",
  "count": 3,
  "results": [
    { "name": "TextField",     "library": "material",   "kind": "class", "url": "https://api.flutter.dev/flutter/material/TextField-class.html",     "score": 0.9474 },
    { "name": "TextFormField", "library": "material",   "kind": "class", "url": "https://api.flutter.dev/flutter/material/TextFormField-class.html", "score": 0.7826 },
    { "name": "BitField",      "library": "foundation", "kind": "class", "url": "https://api.flutter.dev/flutter/foundation/BitField-class.html",    "score": 0.6667 }
  ],
  "index_stale": false
}

flutter_mentions(text, max_tokens=4000)

Parses every @flutter_mcp … mention in a text and resolves each one, returning exactly one entry per mention, in document order — a failing mention never removes, duplicates or leaks state into another one.

Mention

Resolved as

@flutter_mcp provider

pub.dev package, latest release

@flutter_mcp provider:^6.0.0

caret constraint → newest published release satisfying it

@flutter_mcp provider:6.1.5

exact published release

@flutter_mcp dio:>=5.0.0 <6.0.0

range → newest published release satisfying it

@flutter_mcp provider:latest

explicit latest

@flutter_mcp pub:dio, pub:dio:5.4.0

explicit package form

@flutter_mcp material.AppBar

Flutter class, library given

@flutter_mcp dart:async.Future

Dart SDK class

@flutter_mcp Container

plain name (index → widgets/material → pub.dev)

Version constraints are answered from pub.dev's own list of published releases. An exact pin that was never published is reported as missing with the nearest published releases — never silently replaced by a different version — and a resolved range always states both the constraint that was requested and the version it resolved to. Pre-releases are excluded unless the constraint mentions one.

Returns {"mentions": n, "max_tokens": …, "results": […]}; {"mentions": 0, …, "note": …} when the text has no mention.

// flutter_mentions({ "text": "Use @flutter_mcp provider:^6.0.0 for state and @flutter_mcp material.AppBar for the bar.", "max_tokens": 120 })
{
  "mentions": 2,
  "max_tokens": 120,
  "results": [
    {
      "mention": "@flutter_mcp provider:^6.0.0",
      "identifier": "provider",
      "requested_constraint": "^6.0.0",
      "type": "pub_package",
      "version": "6.1.5+1",
      "url": "https://pub.dev/api/packages/provider/versions/6.1.5+1",
      "title": "provider 6.1.5+1",
      "content": "A wrapper around InheritedWidget to make them easier to use and more reusable.\n\n\n[truncated: showing ~19 of ~6240 estimated tokens]",
      "truncated": true, "tokens": 19, "source_tokens": 6240, "max_tokens": 120,
      "cached": true,
      "note": "constraint '^6.0.0' resolved to 6.1.5+1: the newest published release satisfying it"
    },
    {
      "mention": "@flutter_mcp material.AppBar",
      "identifier": "material.AppBar",
      "requested_constraint": null,
      "type": "flutter_class",
      "version": null,
      "url": "https://api.flutter.dev/flutter/material/AppBar-class.html",
      "title": "AppBar class",
      "content": "# AppBar class\n\nA Material Design app bar. … [truncated: showing ~80 of ~7367 estimated tokens]",
      "truncated": true, "tokens": 80, "source_tokens": 7367, "max_tokens": 120,
      "cached": true
    }
  ]
}

A mention that resolves nowhere is a per-mention failure, not a missing entry:

// results entry for "@flutter_mcp NotARealThing"
{
  "mention": "@flutter_mcp NotARealThing",
  "identifier": "NotARealThing",
  "type": "not_found",
  "version": null,
  "error": "could not resolve 'NotARealThing': flutter widgets: not found (HTTP 404): …",
  "suggestion": "try flutter_search('NotARealThing') to find the right name"
}

pub_package(package_name, version=None, max_tokens=6000)

pub.dev metadata plus the README. package_name is exact and case-sensitive; version pins a release. Returns {"name", "version", "description", "publisher", "likes", "pub_points", "url", "readme", "truncated", "tokens", "source_tokens", "max_tokens", "cached"} (plus truncation when the README was cut); publisher / likes / pub_points may be null when pub.dev does not report them.

// pub_package({ "package_name": "dio", "max_tokens": 150 })
{
  "name": "dio",
  "version": "5.11.1",
  "description": "A powerful HTTP networking package,\nsupports Interceptors,\nAborting and canceling a request,\nCustom adapters, Transformers, etc.\n",
  "publisher": "flutter.cn",
  "likes": "8.36k",
  "pub_points": 160,
  "url": "https://pub.dev/api/packages/dio",
  "readme": "# dio\n\n… [truncated: showing ~137 of ~7837 estimated tokens]",
  "truncated": true,
  "cached": true
}

flutter_status()

Probes api.flutter.dev and pub.dev with a light GET (10 s timeout) and reports the index and cache state. overall is ok, degraded or error. The politeness block is diagnostics only and never changes overall.

// flutter_status()
{
  "server": "flutter-docs-mcp",
  "version": "0.3.0",
  "checks": {
    "search_index":    { "status": "ok", "entries": 3503, "built_at": "2026-10-06T08:46:45.893682+00:00", "stale": false },
    "cache":           { "status": "ok", "entries": 5, "expired": 0 },
    "api_flutter_dev": { "status": "ok", "http_status": 200 },
    "pub_dev":         { "status": "ok", "http_status": 200 }
  },
  "overall": "ok",
  "politeness": {
    "status": "ok", "disabled": false,
    "requests": 3, "robots_requests": 2, "robots_rows": 2, "robots_fetches": 2, "robots_cache_hits": 1,
    "blocked_by_robots": 0, "throttle_waits": 3, "throttle_sleep_s": 1.203,
    "host_delays": { "api.flutter.dev": 0.0, "pub.dev": 0.0 },
    "budgets": { "fetch:api.flutter.dev": [2, 6] },
    "budget_denied": 0, "conditional": 0, "revalidated_304": 0,
    "retries_429": 0, "retries_transport": 0, "stalls": 0, "errors": 0
  }
}

health_check()

{"status": "ok", "server": "flutter-docs-mcp", "version": "0.3.0"} — no network, no cache; safe as a liveness probe.

Replacing the legacy flutter-mcp-server

This server exists so the third-party flutter-mcp-server (0.1.1, upstream flutter-mcp/flutter-mcp) can be removed from our MCP client configuration. We do not maintain or patch that package; the three defects below are what we observed in its code while checking whether it could cover our use case, and they are the reason we wrote this one instead:

  1. flutter_docs crashes on the pub.dev path. server.py:1562 calls create_truncator(tokens), but that package's truncation.py:250 defines create_truncator() with no parameters, so any call that passes tokens fails with Error executing tool flutter_docs: create_truncator() takes 0 positional arguments but 1 was given. The same block then calls a truncator.truncate(content) method that does not exist.

  2. process_flutter_mentions invents results. Its version-constrained pub branch does not continue after resolving the package, so execution falls through into the shared "process the result from flutter_docs" block with the previous iteration's doc_result still bound. The output contains a phantom second entry (type: flutter_widget, empty content) for a mention that was never in the text.

  3. Its Flutter-class path needs a writable Flutter SDK. It shells out to a local /opt/flutter checkout, whose bin/cache is read-only on our hosts, so Flutter class lookups fail there. This server scrapes api.flutter.dev and api.dart.dev instead and needs no Flutter installation.

flutter_mentions is the replacement for process_flutter_mentions: same mention grammar, one result per mention, and version constraints resolved against pub.dev's published release list. The truncation contract is also stricter here — truncated, tokens and source_tokens describe the payload that was actually returned (see flutter_docs).

Caching & politeness

Cache

Everything lives in one directory: ~/.cache/flutter-docs-mcp by default, overridable with FLUTTER_DOCS_MCP_CACHE_DIR.

File

Contents

TTL

cache.db

fetched pages (parsed result + raw body + ETag / Last-Modified), the search index, pub.dev data

docs 7 days, pub.dev 1 day, index 7 days

robots.db

the politeness layer's robots.txt cache

7 days per host

Delete the files to force a full refresh. A cache problem is never a tool failure: if the directory is unwritable the server simply runs without a cache and says so in flutter_status().checks.cache.

Politeness

Every outbound request — fetch_*, the index build and the flutter_status probes — goes through one small internal module, src/flutter_docs_mcp/politeness.py (stdlib + httpx, no extra dependency):

  • robots.txt is read and obeyed. Rules are matched per RFC 9309 (*, trailing $, %2A/%24 literals, merged User-agent groups, most-specific match wins, Allow beats Disallow on a tie). urllib.robotparser is deliberately not used: on Python < 3.14 it ignores wildcards and would report rules such as Disallow: /pypi/*/json as allowed. A disallowed URL returns {"error": "… blocked by robots.txt …"} — no request is made and no exception escapes. Files are cached 7 days per host, including negative results (404 / 403 / 5xx), so a host costs one robots request per week.

  • Per-host throttle, including Crawl-delay. Requests to one host are spaced 0.35–0.9 s apart, or by the site's own Crawl-delay when it declares one. A cold index build is ~31 requests, which is exactly where that delay is felt — that is the point.

  • 429 / 503 / 504 are handled. Retry-After is honoured to the second; without it the per-host delay is escalated and the request is retried once. A response that stalls past the read timeout escalates the delay instead of being retried blindly.

  • Conditional GET. ETag / Last-Modified and the raw body are stored next to the cached result, so refreshing an expired entry costs a 304 instead of a full download.

  • Request budgets. One fetch_* call may make at most 6 requests per host — the robots fetch, every retry and every redirect hop all pay. A cold pub_package legitimately costs 3 (robots + API JSON + package page). An index build gets 40 layer calls per host; it charges one unit per page and the layer does robots and any hop inside it (measured cold build: 24 units on api.flutter.dev, 7 on api.dart.dev).

  • Host allowlist. Only api.flutter.dev, api.dart.dev and pub.dev can be contacted, so a malformed identifier or a surprising redirect cannot turn a docs lookup into a request to somebody else's site.

The layer never raises and never changes a tool's return shape. flutter_status reports its counters under the top-level politeness key.

Opt-out

export FLUTTER_DOCS_MCP_POLITENESS_DISABLED=1

This single switch turns off robots.txt, the throttle, retries and conditional GET at once. Use it at your own risk: you take over responsibility for respecting each site's crawling rules, and you are far more likely to be rate-limited or blocked. There is no partial opt-out.

Development

python3 -m venv .venv
.venv/bin/pip install -e ".[dev]"

.venv/bin/python -m pytest -q              # 255 offline tests; fixtures in tests/fixtures/
.venv/bin/python scripts/politeness_smoke.py /tmp/flutter-docs-smoke-cache
.venv/bin/python scripts/e2e_mcp_test.py
  • pytest is fully offline: HTTP is simulated with httpx.MockTransport and time/jitter are injected, so the suite is deterministic and runs in seconds.

  • scripts/politeness_smoke.py <cache-dir> is a live measurement, not a test: it drives the real tools against the real docs sites and prints what the politeness layer did (requests, robots, throttle, conditional GET, budgets).

  • scripts/e2e_mcp_test.py spawns the installed flutter-docs console script and speaks newline-delimited JSON-RPC to it (initialize → tools/list → every tool → a negative case → health_check). Exit code 0 means every check passed. Set FLUTTER_DOCS_MCP_E2E_CMD to run a different server command.

Troubleshooting

The first flutter_docs call takes half a minute. That is the cold search-index build: ~47 page fetches across api.flutter.dev (1 index page + 25 libraries) and api.dart.dev (1 index page + 20 libraries), spaced by the per-host throttle (a measured cold run finished in ~26-30 s). It happens once every 7 days; later calls hit the cached index. Delete cache.db and you pay for it again.

request budget exhausted for scope 'fetch:pub.dev'. One fetch_* call hit its cap of 6 requests to that host. Redirect chains, 429 retries and transport retries all consume the budget, so this usually means the site redirected more than expected or the connection kept failing. The tool returns an error dict instead of hammering the host; flutter_status().politeness.budgets shows how much each scope spent.

"partial": true, or fewer search results than expected. An index build ran out of its 40-calls-per-host budget and stopped early, returning a partial index with a partial_reason such as request budget of 40 per host exhausted for index:api.flutter.dev; the index is incomplete. A partial index is not cached, so the next run rebuilds it. flutter_search still works — it just knows fewer classes — and flutter_status().checks.search_index.entries tells you how many it has.

"stale": true / index_stale: true. The rebuild failed (offline, DNS failure, 5xx) and the server fell back to the older cached index instead of failing the lookup.

Nothing is cached and overall is degraded. Read flutter_status().checks.cache.error — an unwritable FLUTTER_DOCS_MCP_CACHE_DIR (read-only mount, missing permission) is the usual cause. Tools keep working without a cache; they are just slower and noisier on the network.

A lookup returns blocked by robots.txt. The site's rules disallow that path for this user agent, and the request was not sent. Fetch the page yourself, or accept the consequences of the opt-out switch above.

Support development

flutter-mcp is built and maintained by Michal in his spare time. If it saves you time or makes your team's docs easier to work with, a coffee (or more) would mean a lot — every contribution helps keep the project moving. 🙏

Pay via Revolut: revolut.me/michal4zvc

License & attribution

MIT — see LICENSE. Copyright (c) 2026 Michal Gaspierik.

The design of the politeness layer was inspired by Crawl4AI (© Unclecode, Apache-2.0): a TTL-cached robots store, the wildcard rule translation, and a per-domain rate limiter with escalating backoff. The implementation is a clean-room rewrite in this project's own synchronous stdlib-plus-httpx style — no line was transcribed, translated or mechanically adapted from Crawl4AI, and Crawl4AI is not a dependency of this package. Three defects of the original design are fixed (robots fetched_at refresh, negative-result caching, Crawl-delay support). The GPL-3.0 part of Crawl4AI — its vendored html2text fork — is deliberately excluded: no code, data or dependency from that tree is used or shipped here. The full statement is in NOTICE.

This project is also a clean-room replacement for adamsmaka/flutter-mcp, whose main branch pins mcp to a 2.x development commit while importing the 1.x-only FastMCP, and which recurses between its "unified" tool and the deprecated tool it delegates to. The useful part of that project — fetching real docs — is kept here as fetchers → parsers → cache → search index → tools, with no circular delegation, a pinned 1.x SDK, error-dict failures and an offline test suite.

Part of the keepee self-hosted family. Also see Wiki4AI — a self-hosted collaborative wiki for teams and AI agents, with hybrid semantic search and an MCP server of its own — and the full project index.

Available Tools

6 tools
flutter_docsA

Fetch one Flutter / Dart SDK / pub.dev documentation page as clean markdown.

Identifier forms (resolved in this order):

  • "pub:dio" or "pub:dio:5.4.0" → pub.dev package (description + README)

  • "dart:async.Future" → Dart SDK class (api.dart.dev)

  • "material.AppBar" / "widgets.ListView" → Flutter class with explicit library

  • "ListView" → exact match in the local search index, then a widgets/material fallback, then pub.dev as a last resort

Args: identifier: one of the forms above. topic: optional section filter (e.g. "methods", "properties", "examples", "constructors") — keeps only the matching section plus the class title/description; if no section matches, the full content is returned with a "note" listing the available section headings. max_tokens: rough token budget for the returned markdown (1 token ≈ 4 chars); longer output is cut at a line boundary with a trailing note.

Returns {"type", "identifier", "url", "title", "content", "truncated", "tokens", "source_tokens", "max_tokens", "cached"} — plus "truncation" (what was cut and why) whenever "truncated" is true, and an optional "note" — or {"error", "suggestion"} on failure.

Inputs: identifier (required), topic, max_tokens.

flutter_mentionsA

Parse every @flutter_mcp mention in a text and resolve each one to its docs.

Mention grammar (identifier forms are the same flutter_docs accepts):

  • "@flutter_mcp provider" pub.dev package, latest release

  • "@flutter_mcp provider:^6.0.0" caret constraint

  • "@flutter_mcp provider:6.1.5" exact published release

  • "@flutter_mcp dio:>=5.0.0 <6.0.0" range, space separated

  • "@flutter_mcp provider:latest" explicit latest

  • "@flutter_mcp pub:dio" / "pub:dio:5.4.0"

  • "@flutter_mcp material.AppBar" Flutter class, library given

  • "@flutter_mcp dart:async.Future" Dart SDK class

  • "@flutter_mcp Container" plain name

A version constraint is answered from pub.dev's own list of published releases: an exact pin is returned when published, a range resolves to the newest release that satisfies it (pre-releases are excluded unless the constraint asks for one), and the constraint that was requested is always reported next to the version that was resolved. A version that does not exist is reported as missing with the nearest published releases — never replaced by a different version.

Args: text: any text containing zero or more @flutter_mcp mentions. max_tokens: rough token budget for each mention's documentation payload (1 token ≈ 4 chars). Non-positive disables truncation.

Returns {"mentions": n, "max_tokens": ..., "results": [...]} with exactly one entry per mention, in document order. Each entry carries "mention" (as written), "type" (flutter_class | dart_class | pub_package | not_found | error), the resolved "identifier" and "version", and the docs payload ("content", "truncated", "tokens", "source_tokens", "max_tokens", plus "truncation" when something was cut). Failures are per-mention {"type": "not_found" | "error", "error", "suggestion"} — one bad mention never removes or duplicates another mention's entry.

Inputs: text (required), max_tokens.

flutter_statusA

Health check: search index state, cache stats, and live API probes.

The cache check gains read_only: true plus read_only_reason when the database cannot be written (P5). It stays an "ok" check on purpose: a cache that only reads is a degraded optimisation, not a sick server, so overall does not change.

Probes api.flutter.dev and pub.dev with a light GET (10s timeout) and reports per-check status plus an overall verdict ("ok" / "degraded" / "error"). Never raises — individual failures only mark that check as error.

Also returns a top-level "politeness" block with the politeness layer's counters: robots cache rows/fetches/hits, requests blocked by robots.txt, throttle waits and per-host delays, conditional GETs / 304 revalidations, 429 and transport retries, request budgets and whether the layer is disabled (FLUTTER_DOCS_MCP_POLITENESS_DISABLED). It is diagnostics only and never affects "overall".

Inputs: no named parameters.

health_checkA

Check that the server is up and report its version.

Inputs: no named parameters.

pub_packageA

Fetch pub.dev package metadata plus its README as markdown.

Args: package_name: exact pub.dev package name (case-sensitive), e.g. "dio". version: optional specific version, e.g. "5.4.0" (default: latest). max_tokens: rough token budget for the returned README (1 token ≈ 4 chars); longer output is cut at a line boundary with a trailing note.

Returns {"name", "version", "description", "publisher", "likes", "pub_points", "url", "readme", "truncated", "tokens", "source_tokens", "max_tokens", "cached"} or {"error", "suggestion"} on failure. truncated is true when the README was cut, in which case a truncation block explains the cut and tokens / source_tokens give the returned / full-page estimates. publisher / likes / pub_points may be null when pub.dev does not report them.

Inputs: package_name (required), version, max_tokens.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.9/5.0

    Scored across 6 tools

    Disambiguation3/5

    Two real overlaps exist: health_check and flutter_status are both described as health checks (up/version vs. index/cache/API probes), and flutter_docs accepts a 'pub:' form that fetches package description+README, duplicating pub_package's core output. Descriptions do clarify the differences (metadata vs. README, version vs. diagnostics), so it's workable but an agent could reasonably pick the wrong one.

    Naming Consistency4/5

    All names are consistently snake_case with no verbs, but the prefixes vary by family (flutter_*, pub_*, plain health_check), which is defensible but not a single predictable pattern. Still readable and mostly uniform.

    Tool Count5/5

    Six tools is well within a healthy scope for a docs server, and each maps to a distinct capability: discovery (flutter_search), fetch (flutter_docs, pub_package), batch resolution (flutter_mentions), and diagnostics (health_check, flutter_status).

    Completeness4/5

    The search-then-fetch workflow, package metadata, and batch mention resolution cover the core docs-retrieval lifecycle well. Minor gaps: no tool to rebuild/refresh the search index or clear the cache despite exposing index_stale and cache state.

    Maintenance

    ActivityMaintained
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      D
      maintenance
      A real-time server that provides Flutter/Dart documentation and pub.dev package information to AI assistants, ensuring they generate accurate and up-to-date Flutter code.
      8
      74
      MIT
    • F
      license
      A
      quality
      D
      maintenance
      Provides AI models with direct access to documentation for over 600 technologies from DevDocs.io, including popular languages, frameworks, and tools. It enables comprehensive searching, content retrieval, and offline access via an intelligent local caching system.
      12
      2
      -
    • F
      license
      B
      quality
      B
      maintenance
      Provides Flutter and Dart reference data, code generation, live API lookups, and a verified sample corpus for AI coding assistants, enabling efficient Flutter development with accurate guidance and boilerplate generation.
      14
      -
    • A
      license
      Not graded
      quality
      C
      maintenance
      Enables AI coding agents to retrieve live CPython standard-library documentation and PyPI package metadata as clean markdown, with cached, throttled, robots-aware requests and local module search.
      MIT