Packrift MCP MCP Server | MCP Server List

Production MCP (Model Context Protocol) server for Packrift exact-spec packaging procurement. Hero use case: find the right packaging supply for a given item, SKU, or reorder need, then confirm live price, inventory, shipping, and cart handoff.

- Stack: Cloudflare Workers, TypeScript (strict), Hono, Zod, Streamable HTTP transport
- Backing: Shopify Admin GraphQL API (2025-04), storepackrift.myshopify.com
- Endpoint:POST /mcp,GET /mcp(SSE),GET /andGET /start(MCP start page),GET /.well-known/mcp/server-card.json
- Install guide: seellms-install.mdfor remote MCP client configuration.
- Discovery gate: product lookup, reorder, quote, and cart handoff flows are constrained to AI-approved Packrift catalog records where SKU selection is involved.

The public Packrift endpoint is hosted at:

Fast start page for developers, agents, and directory reviewers:

Directory and partner handoffs can use tracked start links without changing the MCP endpoint. Use a lowercase source slug with letters, numbers, and underscores:

https://mcp.packrift.com/r/start/{source}

The start page also renders source-specific copy controls when it receives a source:

https://mcp.packrift.com/start?utm_source={source}

Those handoffs record aggregate install-intent telemetry:/r/config/{source}fetches show up asmcp_tracked_config_fetches,/r/install/{source}/{target}opens show up asmcp_install_intent, and copy controls recordmcp_install_copyby source and target so partner and directory handoffs can be evaluated before downstream cart events appear.

Custom partner, campaign, directory, and agent-workflow slugs are allowed without code changes as long as they match^[a-z0-9_]{2,64}$. Examples:mcpservers_org,agency_partner,browser_agent_demo,newsletter_mcp.

When a directory or agent host wants a source-attributed config file instead of an HTML start page, use:

https://mcp.packrift.com/r/config/{source}

Tracked config fetches are exposed by source inhttps://mcp.packrift.com/ai/mcp-usage-snapshot.json.

For a target-specific tracked install action, use:

https://mcp.packrift.com/r/install/{source}/{target}

Common targets includegeneric_streamable_http,stdio_mcp_remote,claude_code,codex,claude_desktop,cursor_windsurf_vscode, andcline.

MCP clients that support remote HTTP or Streamable HTTP servers can add Packrift with:

{ "mcpServers": { "packrift": { "type": "http", "url": "https://mcp.packrift.com/mcp" } } }

Cline users should use the tracked Cline target, which returns Cline-nativestreamableHttpconfig:

https://mcp.packrift.com/r/install/cline_mcp_marketplace/cline?format=json

Hosts that only accept local stdio MCP commands can use the trackedstdio_mcp_remotetarget. It runsnpx mcp-remoteas a thin bridge and still forwards every call to the hosted Packrift endpoint:

https://mcp.packrift.com/r/install/{source}/stdio_mcp_remote?format=json

Use the hosted endpoint above when possible. It requires no buyer-side API key and exposes the current 15-tool Packrift exact-spec commerce surface.

- Glama hosted connector:https://glama.ai/mcp/connectors/io.github.Packrift/packrift-mcp
- MCP Marketplace listing:https://mcp-marketplace.io/server/io-github-packrift-packrift-mcp
- Official MCP Registry search:https://registry.modelcontextprotocol.io/v0/servers?search=Packrift
- Claude connector submission packet:https://mcp.packrift.com/ai/claude-connector-submission.json
- Agent capture outreach packet:https://mcp.packrift.com/ai/agent-capture-outreach.json
- Source repository:https://github.com/Packrift/packrift-mcp
- Direct Streamable HTTP endpoint:https://mcp.packrift.com/mcp

The hosted Glama connector should remain the primary Glama traffic target. The separate Glama source server listing can use this repository for release and quality checks, but it should still point users at the hosted no-auth MCP endpoint above.

- Dockerfilestarts the Packrift MCP server onPORT=8787.
- glama.jsonis the source-listing maintainer claim file.
- smithery.yamlgives Glama/Smithery-compatible container metadata with an empty config schema, so source-listing release checks do not infer a required buyer token.
- Running the container withoutSHOPIFY_PACKRIFT_TOKENstill exposes MCP discovery:tools/listreturns the current 15-tool surface andresources/listreturns the public AI/MCP resources.
- SHOPIFY_PACKRIFT_TOKENis not required for MCP discovery, directory scanning, or hosted connector use. It is required only for self-hosted live Shopify-backed catalog, pricing, inventory, shipping, and cart tool calls.
- The source listing should not create a separate Packrift CLI or buyer surface; it should release/sync the existing repository and keep the canonical runtime athttps://mcp.packrift.com/mcp.

The public hosted endpoint above is the primary integration path and the canonicalserver.jsonmanifest is remote-only. A container image is still published for local development and environments that explicitly require a self-hosted package-style install surface:

docker pull ghcr.io/packrift/packrift-mcp:latest docker run --rm -p 8787:8787 \ -e SHOPIFY_PACKRIFT_TOKEN=<shopify_admin_api_token> \ ghcr.io/packrift/packrift-mcp:latest

You can also build the image from this repository:

docker build -t packrift-mcp . docker run --rm -p 8787:8787 \ -e SHOPIFY_PACKRIFT_TOKEN=<shopify_admin_api_token> \ packrift-mcp

- PORTdefaults to8787
- SHOPIFY_STORE_DOMAINdefaults topackrift.myshopify.com
- SHOPIFY_API_VERSIONdefaults to2025-04
- STOREFRONT_DOMAINdefaults topackrift.com
- SHOPIFY_PACKRIFT_TOKENis optional for MCP introspection and directory scanning, but required when self-hosting live Shopify-backed catalog, pricing, inventory, shipping, and cart tools

The container uses an in-memory cache instead of Cloudflare KV. It is intended for discovery and client testing; the production server remains the Cloudflare Workers endpoint athttps://mcp.packrift.com/mcp.

Notify IndexNow about the existing source-specific MCP eval-pack, start, install, first-run, and activation URLs:

npm run submit:source-activation-indexnow

The script preflightshttps://mcp.packrift.com/ai/mcp-source-activation-sitemap.xml, writes artifacts underoutputs/source-activation-indexnow/, and submits only existing hosted MCP URLs. It does not create a new CLI, buyer surface, checkout path, or directory submission.

Tools are framed around exact-spec procurement, not generic browsing. Usefind_packaging_for_itemwhen the buyer has item dimensions or a fit question; use SKU and exact-spec tools when the buyer is replenishing a known product.

The server advertises prompts and resources in addition to tools.

- find_exact_packaging_spec
- fit_item_then_prepare_cart
- review_cart_handoff_candidates
- reorder_packrift_sku
- prepare_cart_handoff
- request_bulk_quote_for_no_match
- copy_procurement_spec
- find_box_by_lwh
- find_label_by_size_material_printer

Resources include the AI-approved product corpus, per-SKU markdown/JSON records, purchase-path JSONL, MCP cart handoff candidates, MCP cart activation playbook, measured handoff directories with cart routes, exact-spec family files, no-exact-match policy, crawler-safe purchase paths, Shopify-native UCP commerce surface, UCP starter catalog, storefront builder kit, and flat storefront import feed for curated packaging/shipping-supplies shelves.

Packrift is optimized for exact-spec packaging procurement: agents should not treat nearby sizes, materials, colors, adhesives, strength ratings, or case counts as exact substitutes. If there is no exact AI-approved match, route the buyer to reorder, no-match explanation, or bulk quote recovery instead of forcing a substitute.

Current paid ChatGPT / product-card order families that are represented in the public corpus:

- SKU1066: 10 x 6 x 6 ECT-32 kraft corrugated boxes, 25/bundle
- SKUMFL1295: 12 1/8 x 9 1/4 x 5 white self-seal literature mailers, 50/pack
- SKULL251WR: 2 5/8 x 1 weather-resistant polyester laser labels, 3000/case

Use/Users/farhan/Downloads/packrift-mcpas the canonical production worktree. The older/Users/farhan/Downloads/packrift-mcp-servercheckout is a legacy duplicate and must not be used for Worker deploys because it can overwritemcp.packrift.comwith stale routes/tools.

Local secrets belong in.dev.vars(gitignored). Do not paste real tokens into this README, issues, directory submissions, screenshots, or third-party support threads.

SHOPIFY_PACKRIFT_TOKEN=<shopify_admin_api_token>

# initialize curl -s -X POST http://127.0.0.1:8787/mcp \ -H 'Content-Type: application/json' \ -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"curl","version":"1"}}}' # list tools curl -s -X POST http://127.0.0.1:8787/mcp \ -H 'Content-Type: application/json' \ -d '{"jsonrpc":"2.0","id":2,"method":"tools/list"}' # search curl -s -X POST http://127.0.0.1:8787/mcp \ -H 'Content-Type: application/json' \ -d '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"search_products","arguments":{"query":"poly mailer","limit":3}}}'

Smoke-test the hosted cart handoff path without creating an order or collecting payment:

This synthetic check calls the current hosted MCP endpoint, verifiesget_cart_handoff_candidates -> get_product -> get_pricing -> check_inventory -> create_cart_url, and checks the MCP cart landing route. It writes JSON and Markdown evidence underoutputs/mcp-cart-handoff-smoke/. Add-- --sku LL251WR --qty 1to test another AI-approved SKU. Use-- --verify-final-cartonly when intentionally checking Shopify's final cart redirect.

GitHub Actions also runs the hosted smoke checks every six hours and on manual dispatch through.github/workflows/mcp-live-smoke.yml. The scheduled job checks both1066andLL251WR, then uploads smoke and capture evidence artifacts. Runnpm run check:distributionfor the broader directory and registry monitor; it exits nonzero only when a tracked surface has a truefailstatus.

Refresh thellms-full.txtpriority SKU block from recent GA4 item activity joined to the AI-approved catalog and live Shopify inventory:

npm run build:llms-full -- \ --ga4-items /path/to/packrift-ga4-items.csv \ --approved-jsonl /path/to/packrift-ai-approved-products.jsonl \ --limit 20

The generated public block must not include local paths, internal notes, or non-public campaign details. A private JSON report is written underoutputs/llms-full-priority-skus/.

To run the whole refresh path, pull fresh GA4 reports, publish the publicllms-full.txtoverride into Workers KV, purge the public agent URLs, and verify the live surfaces:

npm run refresh:llms-full -- --publish-kv

This updates thestatic-override:llms-full.txtKV key. The Worker serves that override after the current deploy, so future priority-SKU refreshes do not require deploying unrelated Worker source changes.

Check crawler-facingllms-full.txtavailability across six AI user agents:

npm run check:static-availability -- --samples-per-ua 200 --concurrency 30

This writes JSON and Markdown evidence underoutputs/static-availability/. The funnel snapshot also runs this check and reports failure rate, 5xx rate, content validation, and p95 latency.

Refresh the full MCP funnel proof, including first-party Shopify order stitching for MCP cart attributes:

The snapshot reads the token-protectedhttps://mcp.packrift.com/admin/mcp-ordersendpoint, scans recent Shopify orders for MCP cart attributes such aspackrift_mcp_key,packrift_ai_id, andutm_source=chatgpt-mcp, and reports attributed order count and revenue beside GA4 cart-landing evidence. The admin endpoint requiresMCP_STATS_TOKENand does not expose customer PII.

Publish the sanitized GA4 visitor proof from that local snapshot to the public Worker surface:

npm run publish:ga4-funnel-proof -- --publish-kv

This updateshttps://mcp.packrift.com/ai/mcp-ga4-funnel-proof.jsonand.mdwithout exposing local paths, raw CSV rows, order rows, or credentials. The public funnel snapshot also reads this proof for the thousands-of-qualified- visitors gate when it is available.

npm run check:distribution npm run build:directory-submission-pack

The distribution check reports which public MCP directories are current, stale, blocked, or failing. The submission pack turns stale rows into copy-ready listing fields, live proof URLs, and the next refresh action for each directory.

npm run check:agent-capture npm run build:agent-capture-outreach

…

Packrift MCP

About

Details

Jump to

Setup

About

Details

Jump to

Setup

More like this

@mendableai/firecrawl Mcp Server

Umut Naci

Playwright

Sequential Thinking

Filesystem

EverArt