Feature Library

Measured WebGPU and web UI parts with guides and pitfalls. Plan free; install licensed parts.

Should I use this

Quality & Safety

A
Description quality
100%
Schema completeness
90%
Naming quality
98%
Poisoning risk
100%
Permission match
100%
Protocol compliance
100%

Based on automated analysis of tool definitions and protocol compliance.

Context Cost

~2,626Tokens (tool definitions)
~1.3 KBTypical response size
Significant attention impact (2.05% of 128k context)

This is the approximate number of tokens consumed each time the server's tools are loaded into a model's context. Higher counts reduce the attention available for other tasks.

Install

One-Click Install

Add this to your `claude_desktop_config.json` file:

{
  "mcpServers": {
    "parts": {
      "url": "https://featurelibrary.com/mcp"
    }
  }
}

Remote endpoints

https://featurelibrary.com/mcpstreamable-http

What it can do

Tool inventory

Tools (11)

🟢 Read-only🟡 Write🔴 Delete⚪ Unknown
🟢find_parts(job, platform, budget, limit, kit)

Free. Ranks the library’s parts for a job described in words, with why each fits and what it costs on the target platform (median ms a frame at 1920×1080 on a measured machine and tier). Strong matches for the job come first (each part says match: strong or loose), then looser ones; within each, parts that fit the platform and budget lead, then those over the budget (overBudget), then those that need WebGPU on a platform without it (blocked). For an interface job, 3D, simulation and effects parts rank lower. Each part says what it runs in (runsIn): raw WebGPU passes, a three.js scene, a 2D canvas or the DOM; a three.js part needs a three.js app, so check it against your renderer before installing. When the words name a look or ask for one, the matching looks come back too. Search by the job, not by a name.

Input Schema

{
  "type": "object",
  "properties": {
    "job": {
      "type": "string",
      "description": "What the part has to do, e.g. \"date range filter for a dashboard\" or \"grass that bends round a walker\"."
    },
    "platform": {
      "type": "string",
      "description": "The target: desktop (the reference laptop, the default), laptop (integrated GPU), phone, a machine id from a part’s costs, or \"no-webgpu\" for browsers without it."
    },
    "budget": {
      "type": "number",
      "description": "Ms a frame the part may cost on that platform (the worse of GPU and CPU)."
    },
    "limit": {
      "type": "integer",
      "minimum": 1,
      "maximum": 30,
      "description": "At most this many parts (default 10)."
    },
    "kit": {
      "type": "string",
      "description": "Optional: a kit id (e.g. dashboard) to search only its parts and those of the kits it is built on."
    }
  },
  "required": [
    "job"
  ],
  "additionalProperties": false
}
🟢get_kit(purpose, tone, platform, level)

Free. A tested set of parts for a kind of build (by kit id, or matched to the purpose in words), built on the foundations, with a tone applied when given: the parts with why, role, stack, needs and costs, the optional candidates, the glue you build yourself, what the library lacks, the tone’s settings and per-part knobs, and the looks that fit, at the level asked for (exact, tuned or inspired: what may change, what must stay); and the page shapes the purpose calls for (starting points for the page as a whole), or, when none fits, how to compose one.

Input Schema

{
  "type": "object",
  "properties": {
    "purpose": {
      "type": "string",
      "description": "A kit id, or the kind of build in words, e.g. \"a calm dashboard for finance\"."
    },
    "tone": {
      "type": "string",
      "description": "Optional tone id from the catalogue (e.g. strict, refined, playful)."
    },
    "platform": {
      "type": "string",
      "description": "Optional target platform for the costs shown, as for find_parts."
    },
    "level": {
      "type": "string",
      "enum": [
        "exact",
        "tuned",
        "inspired"
      ],
      "description": "How closely the build follows a look: exact (the look as shipped; the dials and switches only, the default), tuned (also the look’s knobs, in range, then check_look) or inspired (your own look carrying its character; never its name)."
    }
  },
  "required": [
    "purpose"
  ],
  "additionalProperties": false
}
🟢get_part(id)

The free card always (summary, role, needs, costs, public API signatures, pitfall titles, pairs-with). With a licence token that owns the part: the licensed card too, with every pitfall in full, the guide, documented signatures, the module’s files and the changelog. Read every pitfall before building with it.

Input Schema

{
  "type": "object",
  "properties": {
    "id": {
      "type": "string",
      "description": "The part id (an old id resolves to the renamed part)."
    }
  },
  "required": [
    "id"
  ],
  "additionalProperties": false
}
🟢get_recipe(scenario, tone, platform, level)

Licensed (needs a licence token). The wiring for a scenario: the kit that fits, the steps in order (tier, install, tokens, wiring between the parts, glue, what to build yourself, tone, checks), and for each part you own its pitfalls in full and its guide. Parts you don’t own are listed with their titles, to buy or replace.

Input Schema

{
  "type": "object",
  "properties": {
    "scenario": {
      "type": "string",
      "description": "The build in words, or a kit id."
    },
    "tone": {
      "type": "string",
      "description": "Optional tone id."
    },
    "platform": {
      "type": "string",
      "description": "Optional target platform, as for find_parts."
    },
    "level": {
      "type": "string",
      "enum": [
        "exact",
        "tuned",
        "inspired"
      ],
      "description": "How closely the build follows a look: exact (the look as shipped; the dials and switches only, the default), tuned (also the look’s knobs, in range, then check_look) or inspired (your own look carrying its character; never its name)."
    }
  },
  "required": [
    "scenario"
  ],
  "additionalProperties": false
}
🟢install(id, version, path)

Licensed (needs a licence token that owns the part). Returns the part’s module/ files, each marked with the licence and this fetch, as resources to write unchanged into a folder of their own; plus NOTICE.md where there is one and the licence. Pin a version to be sure of what you get. Also takes a look id (e.g. splash, with a whole-library licence): the look’s stylesheet, its backdrop where it has one, the shared dials and the plain tokens, and the order to load them in.

Input Schema

{
  "type": "object",
  "properties": {
    "id": {
      "type": "string",
      "description": "The part id, or a look id."
    },
    "version": {
      "type": "string",
      "description": "Optional: the version to install (only the current one is served)."
    },
    "path": {
      "type": "string",
      "description": "Optional: one file of it, as listed in files (e.g. module/charts.ts): returned as plain text, marked, with its SHA-256. For clients that cannot show the files as resources, or a reply too large to read at once."
    }
  },
  "required": [
    "id"
  ],
  "additionalProperties": false
}
🟢get_playbook(id)

Free. The library’s playbooks: how parts are placed, moved, made accessible and kept fast, with the reason for every rule and a checklist. Without an id, the list (title, summary, the parts each covers); with an id, the playbook in full as markdown. get_kit and get_recipe name the playbooks that cover their parts: read those (playbooks) before wiring the parts; the ones under playbooksIfUsed only when the build uses that part or does that work.

Input Schema

{
  "type": "object",
  "properties": {
    "id": {
      "type": "string",
      "description": "Optional: the playbook id, e.g. placement or accessibility."
    }
  },
  "additionalProperties": false
}
🟢check_look(look, level, mode, dials, switches, ...)

Free. Runs the library’s own checker on how a build uses a look: each override is one of the look’s knobs, of its kind and in range; each switch is one of its options; and text contrast holds (WCAG AA) in each mode at the dials’ extremes with the overrides applied. Returns ok, the problems in sentences, the label the build may use ("In Splash", "Splash, tuned"; none for inspired) and, for a tuned build that passes, the stylesheet to load after the look’s.

Input Schema

{
  "type": "object",
  "properties": {
    "look": {
      "type": "string",
      "description": "The look id, e.g. splash."
    },
    "level": {
      "type": "string",
      "enum": [
        "exact",
        "tuned",
        "inspired"
      ],
      "description": "Optional: tuned when overrides are given, otherwise exact."
    },
    "mode": {
      "type": "string",
      "enum": [
        "light",
        "dark"
      ],
      "description": "Optional: check one mode; every mode the look has when left out."
    },
    "dials": {
      "type": "object",
      "description": "Optional: { liveliness, fidelity, density }, each 0 to 100; the look’s own for any left out.",
      "additionalProperties": {
        "type": "number"
      }
    },
    "switches": {
      "type": "object",
      "description": "Optional: switch values by attribute, e.g. { \"data-splash-drips\": \"liquid\" }.",
      "additionalProperties": {
        "type": "string"
      }
    },
    "overrides": {
      "type": "object",
      "description": "Optional: knob values, { \"--accent\": \"#0f6e66\" } for every mode, or per mode { \"light\": {…}, \"dark\": {…} }."
    }
  },
  "required": [
    "look"
  ],
  "additionalProperties": false
}
🟢check(evidence, decisions)

Free. The rules version (its hash) and the catalogue version, the checks to run on the build before calling it done, the runnable checks (get_check) that prove the ones a checklist can’t, and, with a licence token, the account’s licences and the parts it owns. Pass evidence (what the checks returned, area by area) and it says whether the build may be called done, and what is missing.

Input Schema

{
  "type": "object",
  "properties": {
    "evidence": {
      "type": "object",
      "description": "Optional, before calling a build done: { layout: [layoutProbe.steps/toggle results], rings: [layoutProbe.rings results], motion: { reducedMotion: true, how }, performance: { msPerUpdate, target }, placement: { widths: [1440, 375] } }. Only layout, rings and motion (they protect people) can fail a build; the rest is worth a look."
    },
    "decisions": {
      "type": "array",
      "description": "Optional: the library defaults this build broke, as design decisions: [{ rule, chose, why }]. The person’s wishes win over defaults; say why for each.",
      "items": {
        "type": "object",
        "properties": {
          "rule": {
            "type": "string"
          },
          "chose": {
            "type": "string"
          },
          "why": {
            "type": "string"
          }
        }
      }
    }
  },
  "additionalProperties": false
}
🟢get_check(id)

Free. Scripts you run on your own build, in its page, for what a checklist can’t prove: layout-probe measures that nothing moves when the view changes state (column edges, row tops, controls that come and go) and that every focus ring shows whole. Without an id, the list with what each proves and how to run it; with an id, the script itself as text. Run them at every width you claim and pass the results to check({ evidence }).

Input Schema

{
  "type": "object",
  "properties": {
    "id": {
      "type": "string",
      "description": "Optional: the check id, e.g. layout-probe."
    }
  },
  "additionalProperties": false
}
🟢get_shape(id)

Free. The page shapes: starting points for a whole page by kind (app workspace, dashboard, product reveal, service landing, story, personal homepage, event, shop, game), each with its reader’s job, reading path, first screen, header, structure, phone, motion, what fits and what not to do. Without an id, the list; with an id, the shape in full; with "compose", the method for a page no shape fits, with a worked example. A shape is where to start, never a template: pick one or compose, and name it in your notes.

Input Schema

{
  "type": "object",
  "properties": {
    "id": {
      "type": "string",
      "description": "Optional: a shape id, or \"compose\"."
    }
  },
  "additionalProperties": false
}
🟢get_brief

Free. The agent protocol (rules.md) to follow, in order, when building with the library’s parts. Also the resource feature-store://rules.md.

Input Schema

{
  "type": "object",
  "properties": {},
  "additionalProperties": false
}

Community

Rate this Server

Evidence

Recent observations

verifiedversion not recorded11 tools