CrafThinkERA MCP
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@CrafThinkERA MCP Start a project to produce a custom machined aluminum bracket."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Crafty MCP
Give an agent tools for the part between "make this" and a real physical object.
The default runtime is local and BYOK. Your Gemini and Meshy keys stay on your machine and go only to those providers. A CrafThinkERA token is optional until you want to cross into hosted/network operations.
intent
|
v
local project
|
+--> Gemini (your key) -> concept -> separate reference views
|
+--> Meshy (your key) -> GLB
|
v
physical handoff
|
v
CrafThinkERA -> quote -> human approval -> payment -> productionCrafty is not another model. The open-source runtime handles the client edge, local project state, provider adapters and public artifact contracts. CrafThinkERA keeps routing intelligence, commercial state, maker operations and production execution private.
Run it
Once the package is published:
npx -y @crafthinkera/mcp doctorUse your own provider keys:
export GEMINI_API_KEY="..."
export MESHY_API_KEY="..."
npx -y @crafthinkera/mcpThe process speaks MCP over stdio, so normally an MCP host launches it for you. doctor is the human-readable sanity check. For Codex, use examples/codex.byok.toml; it forwards the key names from your local environment instead of writing secret values into config.
For example, a stdio host can run:
{
"mcpServers": {
"crafty": {
"command": "npx",
"args": ["-y", "@crafthinkera/mcp"],
"env": {
"GEMINI_API_KEY": "${GEMINI_API_KEY}",
"MESHY_API_KEY": "${MESHY_API_KEY}"
}
}
}
}Related MCP server: Orchestrator Python MCP Server
Local tools
The v0.3 local runtime exposes:
crafty_status— provider readiness, never secret values.start_project— durable local project state under~/.crafthinkera.inspect_project— artifacts, checksums and provider jobs.generate_concept— one canonical image through your Gemini key.generate_reference_views— four separate identity-consistent views; never a contact sheet.generate_geometry— start Meshy image-to-3D with your Meshy key.check_geometry— poll once and save the GLB locally when ready.check_crafthinkera_connection— optional read-only check of a hosted bearer token.
Do not hardcode this list in an agent. Use MCP tools/list at runtime.
Why separate views matter
The geometry input contract is intentionally boring:
front.jpg
left.jpg
back.jpg
right.jpgEach file contains exactly one view of the same object. A contact sheet can make a 3D model interpret the sheet as a scene containing several objects. Crafty never feeds a contact sheet into Meshy.
Meshy's API accepts base64 Data URIs, so the local runtime does not need to upload your reference images to CrafThinkERA or a public bucket before geometry generation.
CrafThinkERA network
Local BYOK does not require CRAFTHINKERA_MCP_TOKEN.
When you have a token, you can verify the hosted MCP connection with check_crafthinkera_connection. The existing hosted endpoint remains:
https://www.crafthinkera.com/api/mcpThe final local-artifact -> CrafThinkERA quote import contract is deliberately not faked in v0.3. The hosted commercial Core currently quotes assets already stored inside a CrafThinkERA project. The next network milestone is a content-addressed handoff/upload flow that imports a local manifest + asset hashes, revalidates the files server-side, and only then allows quoting.
That boundary matters: a locally generated GLB is not automatically production-ready.
Hosted bridge compatibility
If you explicitly want the old thin bridge behavior:
export CRAFTHINKERA_MCP_TOKEN="ctk_..."
npx -y @crafthinkera/mcp --remoteIn default mode the token is optional.
Security model
Provider keys are read from environment variables and are never returned by MCP tools or written into project manifests. Local projects use generated IDs and generated filenames; project IDs and asset filenames are validated before filesystem access. Meshy output downloads are restricted to HTTPS Meshy asset hosts.
Physical side effects stay behind CrafThinkERA's private Core. Quote, approval, payment and production are not inferred from local provider success.
Open / private boundary
Open here:
MCP runtime + transport
BYOK provider adapters
local project/artifact format
public tool schemas
client examples
connection diagnosticsPrivate in CrafThinkERA:
provider/routing policy
cost + quality policy
commercial state
maker selection
production operations
customer dataThe split is intentional.
Development
git clone https://github.com/crafthinkera/mcp.git
cd mcp
pnpm install --frozen-lockfile
pnpm check
pnpm test
pnpm buildLocalhost is only for developing the hosted bridge itself; ordinary BYOK users do not need to run a CrafThinkERA web server.
Discovery
Repo https://github.com/crafthinkera/mcp
Remote https://www.crafthinkera.com/api/mcp
Discovery https://www.crafthinkera.com/.well-known/mcp.json
Skill https://www.crafthinkera.com/skills/crafthinkera/SKILL.md
LLMs https://www.crafthinkera.com/llms.txtRegistry identity: com.crafthinkera/crafty-mcp.
License
MIT
Available Tools
8 toolscheck_crafthinkera_connectionARead-onlyIdempotent
Verify the optional hosted CrafThinkERA bearer token and list the remote production/commerce tools. No side effect.
Inputs: no named parameters.
check_geometryBIdempotent
Check a Meshy task once. When it succeeds, download the GLB into the local project.
Inputs: jobId (required), projectId (required).
crafty_statusARead-onlyIdempotent
Show local BYOK provider readiness without revealing API keys.
Inputs: no named parameters.
generate_conceptB
Generate one canonical concept image with the user's Gemini API key.
Inputs: projectId (required).
generate_geometryB
Start a Meshy image-to-3D task from the latest separate reference views using the user's Meshy API key.
Inputs: projectId (required), textured.
generate_reference_viewsA
Generate four separate consistent geometry views from the latest concept using the user's Gemini API key.
Inputs: projectId (required).
inspect_projectCRead-onlyIdempotent
Read a local project, generated artifacts and provider jobs.
Inputs: projectId (required).
start_projectB
Create a local physical-product project. No CrafThinkERA account or token required.
Inputs: intent (required), title.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 8 tools
Each tool maps to a distinct pipeline step: provider readiness, project creation/read, three generation stages (concept, reference views, geometry), and a geometry task poll. The three 'check/status' style tools (crafty_status, check_geometry, check_crafthinkera_connection) could momentarily be confused but their targets (BYOK providers vs Meshy job vs hosted token) are clearly differentiated.
Consistent snake_case verb_noun pattern throughout (start_project, inspect_project, generate_concept, check_geometry, check_crafthinkera_connection). Minor deviation is crafty_status, which is adjective_noun rather than verb_noun, but it remains readable and predictable.
Eight tools is well-scoped for a linear concept-to-geometry pipeline, with each tool earning its place. No redundant or filler tools.
The pipeline covers status, project creation/read, and all generation/verification stages, forming a coherent end-to-end flow. Minor gaps exist (no update/delete or list_projects for local projects), but core workflows are covered.
Maintenance
Related MCP Connectors
Add human approval to AI agent workflows through cQnce authorization tools.
Connect, monitor, and control AI agents — tasks, approvals, schedules, and governance.
Build and run visual creative-production workflows from your AI agent.
Design, build and run teams of AI agents with checked handoffs between them.
Related MCP Servers
- AlicenseAqualityDmaintenanceEnables AI agents to create, execute, and analyze workflows on a Conductor instance through natural language.1916Apache 2.0
- FlicenseBqualityBmaintenanceEnables AI coding assistants to run a machine-verified DESIGN→PLAN→EXECUTE→VERIFY→COMPLETE workflow with human approval gates, state integrity checks, and DAG task scheduling.7-

handsforagents-mcpofficial
AlicenseNot gradedqualityBmaintenanceEnables agents to request physical-world tasks such as 3D printing, CNC fabrication, assembly, measurement, testing, on-site verification, and receive-repack-ship services from a human operator, with fixed-price quotes, status tracking, and evidence packages.MIT- AlicenseAqualityCmaintenanceEnables AI agents to obtain permission before acting and verify afterward against destination systems' own records, confirming whether the permitted work actually happened.552 npmMIT