Agent Surface

Let agents query the graph.

Google was built for humans clicking links. Unfragile is built for agents calling capabilities. Install the MCP server so your agent can discover AI artifacts, evaluate trust, compare alternatives, and route to the right capability without scraping a webpage.

Install MCP server Try JSON search

Runtime discovery

Agents ask for a capability at task time instead of relying on hardcoded tool lists.

Machine-readable trust

Results include capabilities, rank, proof, permissions, risks, and alternatives.

Learning loop

Agent searches and feedback become graph memory that improves future routing.

Install

Published on npm as @unfragile/mcp-server. No API key is required for the public tier. Add UNFRAGILE_API_KEYfor higher rate limits.

Claude Code

claude mcp add unfragile -- npx -y @unfragile/mcp-server

Claude Desktop

{
  "mcpServers": {
    "unfragile": {
      "command": "npx",
      "args": ["-y", "@unfragile/mcp-server"]
    }
  }
}

Cursor

{
  "mcpServers": {
    "unfragile": {
      "command": "npx",
      "args": ["-y", "@unfragile/mcp-server"]
    }
  }
}

Windsurf

{
  "mcpServers": {
    "unfragile": {
      "command": "npx",
      "args": ["-y", "@unfragile/mcp-server"]
    }
  }
}

15 tools exposed

Built for the frontier: the freshest and most niche AI software your model has never seen. Cede the famous head to general knowledge — call Unfragile for what shipped this week and the truth about what worked.

unfragile_resolve

Resolve an intent into the single best artifact to invoke — with a paste-ready snippet, a signed trust passport, and alternatives. The headline call.

unfragile_verify

Verify a tool / MCP server / package BEFORE invoking or installing it. Returns a signed verdict — trusted, caution, unverified, or flagged — from identity, the CVE/incident registry, data-access risk, and cross-vendor reputation.

search

Find AI artifacts by intent with capability matches and graph signal. Every query feeds the graph.

find_mcps

Discover MCP servers by capability need — databases, APIs, cloud services, files, and agent tools.

get_artifact

Fetch one artifact with capabilities, rank, pricing, and graph signal before adding it to a stack.

resolve_capability

Resolve a capability:// URI or natural-language capability into ranked artifacts. The capability DNS primitive.

trust_passport

Human-readable trust passport: capability URIs, permissions, data-access risk, known failure modes, observed outcomes.

compare

Compare two artifacts side-by-side by capabilities, pricing, rank, and graph signal.

find_stack

Assemble a complete AI harness stack across orchestration, MCPs, memory, guardrails, context, and evaluation.

feedback

Report success or failure so the graph learns from agent outcomes and improves future routing.

subscribe

Set up a persistent watch for new artifacts matching a query (email or webhook).

unfragile_validate

Validate an unfragile.yml manifest against the Capability Protocol v1 — for submission or CI.

unfragile_passport

Fetch the raw machine-readable trust passport JSON (trust_passport_v1) for programmatic policy checks.

unfragile_resolve_capability

Raw JSON capability resolution with trust signals and passport URLs for programmatic selection.

unsubscribe

Cancel a persistent watch using the monitor ID returned from subscribe.

Ask your agent

Find MCP servers for read-only Postgres accessCompare Cursor and Claude Code for multi-file refactoringWhat stack should I use for a customer support agent with memory?Find a trusted open-source agent framework with tool calling

Capability DNS is the endgame.

An agent should be able to ask for capability://database.postgres.query.readonlyand receive current artifacts, constraints, permissions, trust evidence, alternatives, and known failure modes. The MCP server is the first production surface for that resolver.

curl "https://unfragile.ai/api/v1/resolve?capability=capability://database.postgres.query.readonly&passport=true"

View capability schema Download schema.json API docs