Component search for the agents building the physical world.

We're building a component search API and MCP server for fast-moving hardware teams. Ask your agent for a part; get exact manufacturer variants, structured published specs and evidence you can check.

Discuss a pilot → See how it works ↓

Exact variantsStructured specsCited sourcesExplicit unknowns

Result · HM17-48-D5-R

Illustrative

① Series datasheet, Rev C, p.2, Table 1

② Variant outline drawing, p.1

▨ Not in any source. Flagged, not guessed.

01 How it works

One request. Exact parts. A source for every value.

Fictional workflow example

01The engineer asks
In the assistant they already use: Claude, Cursor, an in-house agent.
02The agent calls Avrel
Requirements become typed constraints over a consistent schema.
03Exact variants return
Specific manufacturer part numbers, with missing values left visible.
04Every value is traceable
Select any 1 to see its document, revision and page.

Engineer · gantry Z-axis, pilot build 2

Agent → tool call

MCP · avrelavrel.search_parts({
  category: "stepper_motor",
  constraints: {
    frame:            "NEMA17",
    holding_torque_nm: { gte: 0.45 },
    rated_current_a:   { lte: 2.0 },
    body_length_mm:    { lte: 48 },
    front_shaft:       { diameter_mm: 5, profile: "D-cut" },
    rear_shaft:        "preferred"
  },
  include: ["evidence", "unknowns"]
})

Avrel → structured result

3 variants checked2 meet all hard constraints1 excluded example2 unknowns flagged

01HM17-48-D5-RHalvard Motion✓ Meets all constraints

Holding torque: 0.52 N·m
Current / phase: 1.7 A
Body length: 48.0 mm
Front shaft: Ø5 D-cut × 20
Rear shaft: Ø5 × 10
Rotor inertia: UNKNOWNNot published for -R variant

02SD-1748-2A-FSennet Drive Co.Meets hard constraints · no rear shaft

Holding torque: 0.55 N·m
Current / phase: 2.0 A
Body length: 48.0 mm
Front shaft: Ø5 flat × 24
Rear shaft: NoneSingle-shaft only
Rotor inertia: 68 g·cm²

03C17-40-15DCorvo ElectromechExcluded · below torque requirement

Holding torque: 0.42 N·m
Current / phase: 1.5 A
Body length: 40.0 mm
Front shaft: Ø5 D-cut × 22
Rear shaft: CONFLICTCatalog vs datasheet
Rotor inertia: 54 g·cm²

Agent → engineer

Two parts meet every hard constraint. HM17-48-D5-R is the only one with a confirmed rear shaft for your encoder.

Its rotor inertia isn't published. If it matters for your driver tuning, ask Halvard before ordering samples. The excluded example, C17-40-15D, is 0.03 N·m under your torque target, and its sources disagree on whether a rear shaft is available.

Suitability, compatibility and qualification stay with your team. Avrel gives you the evidence to make that call.

02 What changes

Less time on catalogs. More time on the build.

We're building Avrel for robotics and automation startups moving from prototypes to pilot builds. The aim: less cross-referencing, inside the tools your engineers already use.

TodayWith Avrel

Crossing catalogs

A dozen tabs, distributor filters that each use different units, and a series page that never says which suffix has the rear shaft.

One request from your existing assistant. Results identify exact manufacturer variants, with category coverage made explicit.

Comparing candidates

A comparison spreadsheet rebuilt for every build, limited to whatever each datasheet chose to publish.

Every candidate in one schema with consistent units. Missing values are marked unknown. They are never estimated.

Engineering review

Someone asks where a number came from, and you go back to re-download the PDF.

Published values link to their source and location, with revisions where available. A reviewer can inspect the evidence.

03 Developer interface

Plugs into the agent you already run.

The hosted MCP server and HTTP API bring component research into your existing assistant. Search parts, inspect a record and check which categories and fields are covered.

search_partsSupported constraints in. Exact variants with evidence and explicit unknowns out. Hard constraints stay hard.
get_partFull structured record for one manufacturer part number.
get_coverageSupported categories, fields and source coverage, so agents know what they can query.

// Hosted pilot · private API key
{
  "mcpServers": {
    "avrel": {
      "url": "https://avrel-api-pilot.onrender.com/mcp",
      "headers": {
        "Authorization": "Bearer <pilot key>"
      }
    }
  }
}

// Requires remote MCP with custom headers.
// Start with get_coverage, then search_parts.

POST https://avrel-api-pilot.onrender.com/v1/search
Authorization: Bearer <pilot key>
Content-Type: application/json

{
  "category": "dc_motor_driver",
  "filters": [
    { "field": "dc_motor_count", "op": "gte",
      "value": 2, "unit": "count" },
    { "field": "logic_3v3", "op": "eq",
      "value": "supported" }
  ],
  "unverified_requirements": [
    "Continuous 1 A per motor at 40 C"
  ]
}

// Response excerpt · full records include evidence
{
  "data_mode": "licensed",
  "status": "partial",
  "scope": {
    "category": "dc_motor_driver",
    "manufacturers": ["adafruit"]
  },
  "unverified_requirements": [
    "Continuous 1 A per motor at 40 C"
  ]
}

// Published filters checked. Continuous current
// at these conditions still needs review.

04 Scope

What we're building first, and what we're not.

The first release focuses on component research: published specifications, exact variants and evidence for engineering review.

General notesUnless otherwise specified
Coverage starts narrow, on purpose.
Design partners help choose the first categories. Stepper motors are one candidate, along with other bought-in electromechanical parts.
Finding comes before comparing.
Exact-part and document retrieval may cover more parts than detailed comparison. Agents will be able to check which fields are supported in each category.
Engineers keep the decision.
Avrel provides evidence. Suitability, compatibility and qualification stay with your team.
It isn't a store.
Purchasing, live inventory and complete supplier coverage may come later. They aren't part of the pilot.

05 Where this goes

Software teams add a dependency in one line. Hardware teams still search for one across browser tabs.

Software dependency

$ npm install three

✓ resolved · versioned · documented

Physical dependency

? NEMA 17 · ≥0.45 N·m · Ø5 D-cut · rear shaft

→ avrel.search_parts(…)

NowAs agents take on more hardware design work, they need reliable access to real components and their published specifications. Research across catalogs and datasheets remains a bottleneck.

NextOur goal is to turn scattered catalogs and datasheets into a search layer agents can query, with structured results, clear coverage and evidence behind published values.

Long termFinding a real component should be as programmable as finding a software library. We want to provide the search infrastructure that helps serious manufacturing teams turn more designs into physical products.

06 Design partners

Bring us the last part that took too long to find.

We're looking for robotics and automation startups to help choose Avrel's first categories and test the first release against how their engineers search today.

01
Share one recent search
Tell us the requirement, what you checked and what you ended up choosing.
02
Agree a focused pilot
Once coverage is confirmed, compare the first release against your current search process for accuracy, evidence and time spent.
03
Shape what gets built
Your builds decide which categories, fields and integrations we work on first.