agentforge-trust-mcp
agentforge-trust-mcp is a proxy providing a single MCP connection to discover, trust-audit, and invoke tools across 100+ MCP servers.
Trust & Audit Tools:
check_trust: Fetch a Trust Score (0–100) for any MCP server (by slug, UUID, or GitHub URL), with per-dimension breakdowns across security scan, code health, behavioral audit, community trust, and EU compliance.evaluate_policy: Evaluate a server against custom trust criteria (minimum scores, required/forbidden badges) to get anallowed: true/falsedecision — useful for gating agent actions.list_trusted: Search the catalog by category and minimum trust threshold, returning up to 25 results sorted by trust score.recommend: Describe a use case in natural language and receive ranked, trust-filtered server recommendations via semantic search.
Action Tools (require API key):
discover: Browse the full MCP server catalog with full-text or semantic search, including trust scores and connection metadata.register_agent: Self-register to obtain an API key for action tools.broadcast_search: Execute a query across multiple servers simultaneously, with aggregated and ranked results.call_tool: Invoke any tool on any cataloged server via a proxy that handles authentication, quotas, and trust gating.
Security: Includes a built-in RCE sanitization layer with input validation, call guards, and anomaly detection.
Integration: Works with Claude Desktop, Cursor, LangGraph, AutoGen, CrewAI, and more — setup in ~3 minutes.
agentforge-trust-mcp
One MCP connection → 100+ MCP servers, with trust audit, discovery, and execution.
AgentForge gives any agent a single connection that exposes the entire catalog of MCP servers — trust-scored, searchable, and executable. Trust Scores (0–100) are computed across five dimensions: code health, security scan, behavioral audit, community trust, and EU compliance.
v0.3.0 ships 8 tools (4 trust + 4 action) plus a P0 RCE sanitization
layer (input validation, call guards, anomaly detection) with 27/27
tests passing and zero external runtime dependencies.
📖 Full integration guide for external multi-agent systems: See agentforge.community/trust for the full integration guide — OpenClaw, Claude Desktop, Cursor, LangGraph, AutoGen, CrewAI in ~3 minutes.
Installation
npx -y agentforge-trust-mcp@latestClaude Desktop / Cursor / OpenClaw config
{
"mcpServers": {
"agentforge": {
"command": "npx",
"args": ["-y", "agentforge-trust-mcp@latest"],
"env": {
"AGENTFORGE_API_URL": "https://agentforge.community",
"AGENTFORGE_API_KEY": "af_agent_REPLACE_AFTER_REGISTRATION"
}
}
}
}Trust tools (
check_trust,evaluate_policy,list_trusted,recommend) work without an API key. Action tools (broadcast_search,call_tool) require self-registration viaregister_agent(one tool call, no signup).
Environment
Variable | Default | Purpose |
|
| Override for self-hosted |
| none | Enterprise tier (private catalogs, custom policies) |
Related MCP server: Agent Identity MCP Server
Tools
Trust tools (no auth required)
check_trust
Returns the AgentForge Trust Score for a server identified by server_id,
slug, or github_url. Always call this before connecting to a new server.
evaluate_policy
Allow/deny gate. Pass a policy like {min_overall: 70, required_badges: ["actively_maintained"]}
and get back allowed: true|false with per-check detail.
list_trusted
Search the AgentForge catalog for servers matching a category and trust threshold. Useful for "find me a secure database MCP server with overall ≥ 80".
recommend
Natural-language search with trust filter. "I need to validate Czech VAT IDs" returns ranked results filtered by minimum trust.
Action tools (new in v0.2.0)
discover
Browse the entire catalog with full-text or semantic search. Returns ranked results with trust scores, categories, and connection metadata. No API key required — read-only catalog access.
broadcast_search (API key)
Run a single query across N servers in one call. AgentForge fans out the search, aggregates results, and returns a unified ranked list. Pay-per-call billing through your wallet.
call_tool (API key)
Invoke any tool on any server in the catalog through AgentForge's proxy. One connection, hundreds of downstream servers — the proxy handles auth, quotas, and trust gating.
register_agent
Self-register your agent and receive an af_agent_* API key. No human
signup, no dashboard — first tool call returns the key. Pass name,
description, capabilities, optional homepage_url and metadata.
Use cases
Trust audit before connecting
Agent: user wants me to use "some-financial-mcp" server.
Agent: let me check its trust first…
→ check_trust(slug="some-financial-mcp")
← overall_score: 42, badges: [], security_scan: pending
Agent: trust is too low, skipping. Let me find alternatives.
→ recommend(query="invoice conversion Czech VAT", min_overall=75)
← 3 results with scores 87, 84, 79
Agent: connecting to the 87-scored server instead.One connection, 100+ servers (v0.2.0)
Agent: I need flight + weather + currency data for a trip planner.
Agent: register once if I haven't already…
→ register_agent(name="TripPlanner", capabilities=["travel"])
← af_agent_xyz123
Agent: broadcast across travel-tagged servers.
→ broadcast_search(query="flights Prague→Tokyo May 15", category="travel")
← results from 4 servers, all trust ≥ 75
Agent: invoke top result through the proxy.
→ call_tool(server="skyscanner-mcp", tool="search_flights", args={...})
← booking resultsNo need to install or configure individual MCP servers — one connection to AgentForge, and your agent can reach the whole catalog.
Trust dimensions
Dimension | Weight | Source |
Security Scan | 30% | Snyk, GitHub Advisory, secret scanning |
Code Health | 20% | Commit recency, issue ratio, license |
Behavioral Audit | 20% | Claude-powered source review, red flag detection |
Community Trust | 15% | Stars, forks, author reputation |
EU Compliance | 15% | GDPR, AI Act, data residency |
Audits rotate every 14 days; evaluate_policy reports partial: true if not
all dimensions are current.
License
MIT — AgentForge 2026
Available Tools
4 toolscheck_trustA
Fetch the AgentForge Trust Score for an MCP server. Returns the overall score (0-100), per-dimension breakdown (code_health, security_scan, behavioral_audit, community_trust, eu_compliance), and badges. Use before connecting to any MCP server.
Inputs: server_id, slug, github_url.
evaluate_policyA
Check whether a server passes a trust policy. Returns allowed:true/false plus individual check results. Example policy: {min_overall: 70, required_badges: ['actively_maintained']}. Use this to gate agent decisions like 'should I use this server for financial data?'.
Inputs: policy (required), server_id, slug, github_url.
list_trustedA
Search AgentForge catalog for servers matching a category and minimum trust threshold. Returns up to 25 results sorted by trust score.
Inputs: category, min_overall, required_badges, limit.
recommendA
Given a natural-language use case, recommend MCP servers filtered by trust. Example: 'I need to validate Czech VAT IDs and convert ISDOC invoices'. Uses AgentForge semantic search + trust filter.
Inputs: query (required), min_overall, limit.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 4 tools
Each tool has a clearly distinct purpose: check_trust evaluates a server's score, evaluate_policy checks policy compliance, list_trusted searches for servers by criteria, and recommend provides natural-language recommendations. No overlap.
Tool names consistently use imperative verbs (check, evaluate, list, recommend) without nouns, which is a simple pattern. However, 'list_trusted' omits a noun like 'servers' for clarity, and the pattern could be slightly more descriptive, but overall consistent.
With only 4 tools, the scope is focused on trust evaluation and discovery. Each tool addresses a core need: score check, policy evaluation, search, and recommendation. No unnecessary tools, and the count feels appropriate for a trust-assessment server.
The tool surface covers key operations: retrieving trust scores, checking policies, searching, and recommending. Minor gaps might include batch operations or detailed audit logs, but the core functionality for trust assessment is well-covered.
Maintenance
Related MCP Connectors
Independent trust scores, tool surfaces and change history for MCP servers.
MCP server for mandates, delegation, policy-gated execution, credential grants, and audit.
Trust verification for MCP servers. Check scores, scan for security issues, search 4,200+ servers.
Trust-scored search engine for MCP servers. 1,900+ sources indexed. IETF draft published. Referenced by OWASP MCP Security Cheat Sheet. L0-L4 trust levels based on cryptographic verification.
Related MCP Servers
- AlicenseAqualityDmaintenanceAn MCP server that exposes tools for issuing scoped agent credentials, delegating narrower child credentials, handling approvals, revoking task trees, and retrieving audit trails and evidence packets.141Apache 2.0
- AlicenseNot gradedqualityDmaintenanceMCP Server for AI agent identity and authorization. Create, verify, and manage agent identities with trust scores and scoped authorization tokens.31 PyPIMIT
- FlicenseAqualityCmaintenanceReputation and trust scoring service for AI agents, exposed as an MCP server. Evaluate counterparties, report interactions, issue portable trust certificates, and detect Sybil attacks.2370 PyPI-

nanmesh-mcpofficial
AlicenseAqualityCmaintenanceMCP server for NaN Mesh, enabling AI agents to search entities, cast trust reviews, register agents, post content, and query trust scores from the AI trust network.31156 npm1MIT