Pocketlantern

io.github.pocketlanternPublished by GitHub account pocketlanternv0.2.0Updated Oct 24, 2026

Blocker-aware decision layer for AI coding agents, grounded in source-linked, time-sensitive facts.

VerifiedSTDIODesktop onlyKnowledge & MemoryDeveloper Tools

Overview

AI-generated overview

Gives coding agents source-linked decision cards and blocker warnings for technology choices, upgrades, migrations, and licensing questions.

What you can do
PocketLantern is a local MCP server that exposes a search_cards tool, plus get_card and get_related_cards, over a bundled cardset of decision cards covering 27 categories such as auth, frontend, database, ai, serverless, infra, and backend. Each card holds structured facts, tradeoffs, cautions, and reference links, and search results can include blocker warnings drawn from a bundled graph index. Retrieval is keyword-based, with no vector database or embeddings, and the cards ship inside the npm package.
When to use it
Worth adding when an assistant is asked to recommend a vendor, plan a framework or runtime upgrade, choose a database or auth provider, or answer licensing and pricing questions where training data may be stale. It is aimed at coding agents in editors such as Claude Code, Cursor, and Windsurf.
Requirements
Node.js 22 or newer and an MCP client. Installed as the npm package @pocketlantern/mcp-server over stdio, or globally as pocketlantern; the init command registers the server in the client config and adds a project rule line. No account, API key, or network access is needed.
Before you install
The init command writes to the client configuration file and adds a line to the project's agent rules file, so review those changes. Card content is bundled and dated, so facts such as pricing, end-of-life dates, and version requirements may be out of date; verify against the linked official sources before acting.

Installation

In SourceWeft

  1. Open Pocketlantern in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.

Other MCP clients

Follow the launch instructions in the repository.

README

PocketLantern

[npm version] [npm downloads] [CI] [License: MIT]

Your AI agent sounds confident. It missed the blockers.

Prices shift, versions break, licenses change. PocketLantern is a blocker-aware decision layer for AI coding agents, grounded in source-linked, time-sensitive facts.

Try these queries

Ask your AI agent after installing PocketLantern:

"Which auth provider — Clerk, Auth0, or Cognito?"

⚠️ Cognito password hashes are permanently non-exportable⚠️ Auth0 password export requires support ticket — Free tier excluded⚠️ Auth0 Rules/Hooks EOL 2026-11-18 — Actions not portable✅ Clerk has the most flexible migration path

"Should I upgrade to Next.js 16?"

⚠️ Sync API access fully removed — all dynamic calls must be awaited⚠️ Custom webpack config breaks next build⚠️ next lint removed — switch to Biome or ESLint CLI⚠️ Requires Node.js 20.9.0+, TypeScript 5.1.0+, React 19.0+✅ Plan phased migration — codemod doesn't cover webpack or middleware

"Can I use the OpenAI Realtime API?"

⚠️ gpt-4o-realtime-preview removed 2026-05-07⚠️ Beta interface (realtime=v1 header) removed same date⚠️ GA event schema incompatible — 4 event renames✅ Start on GA from day one — or plan dual-stack migration

Three questions. Three blocker types. Same AI, different facts.

[PocketLantern demo — Node.js 20→24 upgrade blockers surfaced in seconds]

Quick start

Prerequisites: Node.js >= 22, an MCP client (Claude Code, Cursor, etc.)

bash
npm install -g pocketlantern

Claude Code

In your project directory:

bash
cd your-projectpocketlantern init

init registers the MCP server in ~/.claude.json and adds a one-line rule to your project's CLAUDE.md. Restart Claude Code (Ctrl+C, then run claude again), then ask your first query.

Cursor / Windsurf / other MCP clients

Add to your MCP config (Cursor: ~/.cursor/mcp.json):

json
{  "mcpServers": {    "pocketlantern": {      "command": "pocketlantern-mcp"    }  }}

Then add this line to your project rules (.cursorrules, Kiro steering file, etc.):

For technology decisions, upgrades, migrations, or licensing questions: before finalizing your answer, check the PocketLantern MCP server's search_cards tool for blockers in these bundled decision cards — your training data may be stale.

Reload MCP servers: Cursor — Cmd+Shift+P -> "MCP: Restart Servers". Windsurf — restart the editor.

More queries to try

  1. "How did Vercel pricing change with Fluid Compute?" Billing split to Active CPU + Memory, shared-process concurrency breaks isolation

  2. "Prisma or Drizzle for Edge/serverless?" Prisma v7 ESM + driver adapter breaking, Edge preview-only; Drizzle 0.30-1.0 also breaking

Verify installation

bash
pocketlantern doctor               # check installation statuspocketlantern search "auth pricing" # search cards from CLI

How it works

PocketLantern is an MCP server. Your AI agent calls search_cards — and gets source-linked decision cards plus source-linked blocker warnings where available.

Agent asks question  -> search_cards("clerk vs auth0 authentication")  -> 3 cards + 5 blocker warnings (lock-in, EOL, export restrictions)  -> Agent answers with source-linked constraints

Local cardset. Instant lookup. No vector database. No embeddings. Keyword search, card links, and source-linked blocker warnings where available. The value is in source-linked decision cards, not clever retrieval alone.

search_cards("auth", constraints:["serverless"]) -> cards + blockersget_card("auth/clerk-vs-auth0-vs-cognito-2026")  -> full card with factsget_related_cards(...)                           -> connected topics

What's included

Blocker-aware decision cards across 27 categories — plus source-linked blocker warnings from the bundled graph index. Ships with the npm package, works immediately, no network, no account:

CategoryWhat it covers
authVendor lock-in, migration pain, SSO, RBAC, passkeys
frontendNext.js, React, Svelte, Vite, Angular upgrade blockers
databasePrisma, Drizzle, Postgres, Aurora, Neon, Supabase
aiOpenAI API migrations, model pricing, batch vs streaming
serverlessVercel, Cloudflare Workers, Lambda cost and runtime
infraHosting platforms, Terraform, Node.js LTS, Python EOL
backendSupabase vs Firebase, Express, job queues, realtime sync
+ 18 moredeployment, devtools, testing, security, compliance, ...

What a card looks like

Each card is a structured YAML file with source-linked facts and official reference links:

yaml
id: auth/clerk-vs-auth0-vs-cognito-2026title: Clerk vs Auth0 vs Cognito Under Current Pricing and Feature Changesproblem: Select an auth vendor given recent pricing shifts, MAU economics, ...constraints: [cost-sensitive, low-ops, enterprise, compliance, serverless]candidates:  - name: Clerk    summary: "Pro starts at $20/mo, includes 50,000 MRUs per app..."    when_to_use: "Choose for small-team + low-ops + cost-sensitive SaaS..."    tradeoffs: "Best DX and fastest implementation..."    cautions: "Be precise about org-member limits..."    links:      - https://clerk.com/pricing      - https://clerk.com/docs/guides/organizations/configure  - name: Auth0    # ...  - name: Amazon Cognito    # ...tags: [auth, clerk, auth0, pricing, passkeys, b2b, compliance]related_cards: [auth/sso-for-b2b-saas, auth/rbac-vs-abac-vs-rebac]updated: 2026-03-14

See packages/schema/src/card.ts for the full schema definition.

Project structure

pocketlantern/  packages/schema/       <- Card schema (zod + TypeScript types)  apps/mcp-server/       <- MCP server (search, retrieval, tool handlers)  apps/cli/              <- CLI (validate, search, init, doctor)  packages/knowledge/    <- Decision cards + graph index (data package)  docs/                  <- User guides & roadmap

Monorepo: pnpm workspace. Build order: schema -> knowledge (no build) -> mcp-server -> cli.

Development

bash
pnpm build              # Build all packagespnpm test               # Run tests (299 tests)pnpm test:coverage      # Run with coverage reportpnpm lint               # ESLintpnpm format:check       # Prettier check

Missing a topic? Request coverage in Card Requests.

Contributing

Contributions to the tool (schema, MCP server, CLI, search) are welcome. See CONTRIBUTING.md for guidelines.

License

MIT

Source: README.md at commit e81973e

Tools

0
SourceWeft has not started this package in its sandbox yet, so its tools are not known.

Version history

1
  1. v0.2.0LatestSep 16, 2026

Score

50% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 days25 pts

    Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.

  • Auth declared10 pts

    Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.

  • Tools listed15 pts

    Answer tools/list without a credential, or ship a package that starts in a sandbox without one.

  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts