Skip to main content

Dependency guardrail for AI coding agents — validates packages before install, catches hallucinations and typosquats. 8,000+ developer tools, compatibility data, migration intelligence.

Project description

IndieStack — Dependency Guardrail for AI Coding Agents

Before your AI installs a package, IndieStack checks if it actually exists, isn't a typosquat, and isn't deprecated. Before it writes infrastructure from scratch, IndieStack finds an existing tool that already does it. 8,000+ packages tracked, npm + PyPI live validation, 422 migration paths from real GitHub data.

Quick Install

Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "indiestack": {
      "command": "uvx",
      "args": ["--from", "indiestack", "indiestack-mcp"]
    }
  }
}

Config file location: ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows)

Cursor / Windsurf / Other MCP Clients

{
  "indiestack": {
    "command": "uvx",
    "args": ["--from", "indiestack", "indiestack-mcp"]
  }
}

Claude Code (CLI)

claude mcp add indiestack -- uvx --from indiestack indiestack-mcp

CLI (terminal)

pip install indiestack

indiestack search "analytics"
indiestack details simple-analytics
indiestack stack "auth, payments, email"
indiestack categories

Both the MCP server and CLI ship in the same package. No API key required.

What it does

AI coding agents hallucinate package names (~20% of recommendations don't exist on npm/PyPI), install deprecated libraries, and write custom code when maintained tools already exist. IndieStack is the guardrail that catches these mistakes before they waste your time.

Install the MCP server and your AI validates packages before every install, gets migration intelligence from real GitHub repos, and finds existing tools instead of generating boilerplate.

Tools (29)

Tool What it does
validate_package PRE-FLIGHT CHECK — verify a package exists on npm/PyPI before installing. Catches hallucinations, typosquats, dead packages. ~500 tokens, saves 10K+
find_tools Search 8,000+ developer tools with 11 filters: price, health, stars, success rate, language, tags, compatibility
get_tool_details Integration code, pricing, API specs, and compatibility data
find_compatible Find tools compatible with a given tool — grouped by category, with verified stacks and conflict warnings
compare_tools Side-by-side comparison of any two tools
build_stack Turn a 50,000-token generation into a 2,000-token assembly
scan_project Analyze a project description + tech stack, get a complete tool recommendation
analyze_dependencies Scan package.json/requirements.txt for better alternatives and health score
evaluate_build_vs_buy Financial breakdown: build from scratch vs use what exists
check_health GitHub health audit — maintenance grade, last commit, stars, alternatives for stale tools
get_migration_data Real migration momentum for any package — how many repos are adopting vs leaving, and what they're switching to/from
get_market_gaps Top unmet needs — what developers search for but can't find. Useful for tool makers deciding what to build.
get_recommendations Personalized suggestions based on your search history
list_categories Browse all 35+ categories with tool counts
list_stacks Curated stacks for common use cases
list_tags All tags sorted by popularity
browse_new_tools Recently added tools with pagination
publish_tool Submit a developer tool so other agents can recommend it
report_outcome Report success/failure after integrating a tool — feeds cross-agent compatibility data. No API key needed.
report_compatibility Report that two tools work well together — builds the compatibility graph
confirm_integration Record a verified integration with notes — strengthens compatibility signals
recommend Record that you recommended a tool — powers Maker Pro citation analytics
shortlist Record which tools you considered — demand signal even for unchosen tools
check_compatibility Check whether a set of tools are compatible with each other
find_agents Search the agent services registry — discover specialist agents for SEO, QA, security, docs, and more
hire_agent Hire an agent service asynchronously — returns a contract ID for inbox retrieval
check_agent_inbox Retrieve completed work from hired agents — delivery type, ref, and summary
rate_agent_delivery Rate whether a delivery was useful — feeds agent quality scores for all future searches
set_api_key Activate an IndieStack API key for higher rate limits and Pro analytics

Resources (3)

Resource What it provides
indiestack://categories All 35+ categories with slugs for filtering
indiestack://trending Top 10 trending developer tools this week
indiestack://tools-index Complete index for prompt caching — include once, reference forever

Prompts (5)

Prompt When to use
before-you-build Check IndieStack before building common functionality
find-alternatives Find indie alternatives to mainstream SaaS products
save-tokens Audit your project for token-saving opportunities
architect-feature Plan a feature using existing indie building blocks
discover-indie Explore what indie developers have built

What's new in v1.19

  • Dependency guardrail — New validate_package(package, ecosystem) tool. Call before every npm install / pip install. Checks live npm/PyPI registries, detects typosquats via edit-distance, flags dead packages, shows migration alternatives. Costs ~500 tokens, saves 10,000+ when it catches a hallucinated install.
  • Guardrail-first instructions — MCP server instructions rewritten. Agents now validate packages before installing, not just search for tools.
  • 29 MCP tools total — Up from 28.

What's new in v1.18

  • Agent-to-agent procurement — 4 new MCP tools: find_agents, hire_agent, check_agent_inbox, rate_agent_delivery. Your coding agent can now discover and hire specialist agents (SEO, QA, security, docs, etc.) for tasks it can't do itself.
  • Async "Claim Check" patternhire_agent returns immediately with a contract ID. The hired agent works asynchronously and delivers results to your inbox. Human gets notified via email/Telegram.
  • Delivery summaries — Hired agents describe what they delivered, so the host agent can explain results to the human without opening the PR.
  • Example payloads — Agent services include example input/output, helping coding agents construct correct payloads without guessing.
  • 28 MCP tools total — Up from 24.

What's new in v1.16

  • Stack compatibility checker — New check_compatibility(tools) tool. Pass 2-8 tool slugs and get a compatibility matrix: which pairs are agent-verified, which are unknown, which conflict. Built for always-on agents doing continuous stack auditing without user prompts.
  • 24 MCP tools total — Up from 23.

What's new in v1.15

  • Migration intelligence via MCP — New get_migration_data() tool. Query real GitHub migration signals for any package: how many repos moved to it, what they moved from, and momentum trend. First time this data moat is exposed via MCP.
  • Description accuracy — Tool count corrected to "8,000+" throughout. PyPI and Registry metadata updated.

What's new in v1.13

  • GitHub stars in results — Search results and tool details now show GitHub star counts. Filter by min_stars to surface popular tools instantly.
  • Smarter category aliases — Searching "cron", "oauth", "smtp", or "uptime" now correctly maps to Scheduling, Authentication, Email, and Monitoring categories.
  • Better alternative exclusions — "[tool] alternatives" queries now exclude wrapper packages (e.g. dj-stripe, laravel-stripe-webhooks) as well as the tool itself.

What's new in v1.12

  • Market gaps — New get_market_gaps() tool exposes zero-result queries ranked by search volume. Agents and makers can see what tools are missing from the ecosystem. API: /api/gaps.
  • Trust-weighted search — Tools with higher agent success rates now rank higher in search results. Real outcome data influences sort order, not just star count.
  • Agent success badges — Search results and explore cards now show agent success rate badges ("93% agent success") when outcome data is available.

What's new in v1.11

  • Migration intelligence — Tool details now include real migration data from 5,000+ GitHub repos. "jest → vitest: 27 repos", "webpack → vite: 13 repos". Agents can recommend tools backed by what developers actually switch to.
  • Verified combos — 60,000+ verified package combinations from production repos. Know what actually works together, not what docs say works together.
  • Unlimited searches — All rate limits removed. Free tier, Pro tier, no limits. Every query is valuable data.
  • Better search relevance — Category-aware scoring. Searching "auth" returns auth tools, not Airflow.
  • 2,100+ install commands — 26% of tools now have install_command populated. Agents can show npm install X immediately.

What was new in v1.9-1.10

  • Smarter ranking — Quality score + GitHub stars ranking.
  • Tech stack filteringscan_project filters by framework compatibility.
  • Super filters — 11 optional filters on find_tools: compatible_with, price, min_success_rate, has_api, language, tags, exclude, health, min_stars, and sort.
  • Compatibility graphfind_compatible returns tools that work together with conflict warnings.
  • Agent outcome trackingreport_outcome with used_with and incompatible_with feeds the compatibility graph.

What was new in v1.7.0

  • Trust tiers — Every tool now shows a trust tier: verified (20+ outcome reports, 70%+ success), tested (5+ reports), or new. Agents can make informed decisions about tool reliability.
  • Agent cards — Machine-readable JSON cards at /cards/{slug}.json with full assembly metadata, health status, and success rates. Index at /cards/index.json. No auth required.
  • Citation milestones — Makers get notified when their tools cross citation thresholds (10, 25, 50, 100, 250, 500, 1000).
  • Trust badges — Tool cards in search results show visual trust indicators based on outcome data.

What was new in v1.6.0

  • Outcome intelligence — Tools now show agent success rates: "82% success rate from 14 agent reports." Report outcomes with report_outcome(slug, success) — no API key needed.
  • Cross-agent intelligence — Every outcome report improves recommendations for all agents across all platforms.
  • Frictionless reporting — Outcome reporting works without an API key, with any scope, and with zero configuration.

What was new in v1.5.0

  • Agent-native actionsrecommend(), shortlist(), report_outcome(), confirm_integration() — agents can now report what they recommend and whether it worked.
  • Scoped API keys — Read or read+write scopes for fine-grained agent permissions.

What was new in v1.4.0

  • CLI toolpip install indiestack now gives you indiestack CLI alongside the MCP server.
  • Pro API enrichment — Pro API keys get richer responses: citation counts, compatible tool pairs, category percentile, and demand context.
  • 3-tier rate limiting — 15/day without a key, 50/day with a free key, 1,000/day with Pro.

What was new in v1.3.0

  • 8,000+ tools — Catalog grew from 830 to 8,000+ via automated GitHub discovery across 37 search queries. Every category now has deep coverage.
  • 500+ compatibility pairs — Auto-generated from shared framework data.
  • README-inferred metadata — Install commands, env vars, SDK packages, and framework compatibility auto-extracted from GitHub READMEs for 700+ tools.

Links

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

indiestack-1.20.0.tar.gz (674.9 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

indiestack-1.20.0-py3-none-any.whl (727.3 kB view details)

Uploaded Python 3

File details

Details for the file indiestack-1.20.0.tar.gz.

File metadata

  • Download URL: indiestack-1.20.0.tar.gz
  • Upload date:
  • Size: 674.9 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.11.2

File hashes

Hashes for indiestack-1.20.0.tar.gz
Algorithm Hash digest
SHA256 769efd6380deabbcac4df769460bcfa171cda9324001f2fee4575d5a9730edd5
MD5 3bc321747ce175d4aad93dc248efe7cd
BLAKE2b-256 a13bc5edef26215c2aa1d85259996d36ddcce58557813fd04a2ab7bb963fd0d3

See more details on using hashes here.

File details

Details for the file indiestack-1.20.0-py3-none-any.whl.

File metadata

  • Download URL: indiestack-1.20.0-py3-none-any.whl
  • Upload date:
  • Size: 727.3 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.11.2

File hashes

Hashes for indiestack-1.20.0-py3-none-any.whl
Algorithm Hash digest
SHA256 fe099765f68ab55b62752bd080737ebc539be122b28568818db9dc257728c5ec
MD5 99140222e98f62c241931d6064da564c
BLAKE2b-256 071a58518d39648b0ed536e3976d947dc3fdb9510a17d8e1474f2971296622f4

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page