{"name":"mcpindex.ai","description":"In-path trust gate for agent tool calls. Pins each MCP tool contract (TOFU) and HOLDs a call before the agent acts when the contract silently changes (a deterministic contract-diff, not a safety verdict), and grades each call's blast radius - what it would do (read, write, delete, send) and whether it can be undone - read from the tool's declared contract (advisory and static, on by default in the clients). Secondary: a public directory indexes MCP servers and publishes per-tool advisory screening verdicts (REVIEW/UNVERIFIED at v1) so an agent can check a tool before it wires it.","version":"1","serversIndexed":31957,"upstream":"https://registry.modelcontextprotocol.io","trust_layer":{"capability":"check_tool_trust","version":"v1-advisory","verdict_contract_version":"1.1.0","verdict_states":["ALLOW","DENY","REVIEW","UNVERIFIED"],"verdict_coverage":{"screened_servers":31478,"unscreened_returns":"UNVERIFIED","screen_conformance":"built_not_yet_run_on_public_corpus_semantic_only","produced_decisions_at_v1":["REVIEW","UNVERIFIED"],"note":"A screened server returns its real REVIEW (semantic-only) from the same getVerdict layer the website renders; an unscreened server returns UNVERIFIED (fail-closed). ALLOW/DENY are reserved in the contract but not produced at v1 - a conforming ALLOW requires the behavioral probe (D3 milestone). Coverage is advisory and semantic-only."},"severity_scale":["INFO","LOW","MEDIUM","HIGH","CRITICAL"],"exposure_tier":"free_no_key_required","methodology":"https://mcpindex.ai/methodology","honest_limits":["conformance_monitored_not_enforced","calibrated_false_v1","advisory_deployment"],"endpoints":["https://mcpindex.ai/api/v1/trust/tool/{server_id}/{tool_name}","https://mcpindex.ai/api/v1/trust/server/{server_id}"],"d3_graduation":{"criterion":"conforming_labels >= 150 AND fp_upper_95 <= 0.02","current_conforming_labels":15,"current_fp_upper_95":null,"status":"pre_graduation","terminal_v1_trigger_date":"2026-09-01"}},"drift_gate":{"what":"in-path trust gate for agent tool calls; pins each tool contract (TOFU) and HOLDs a call before the agent acts when the contract silently changes","method":"deterministic contract-diff over a ChangeKind taxonomy + injection/exfil marker scan (input schema, output schema, description)","change_kinds":["added-required-param","added-optional-param","required-set-expanded","constraint-narrowed","type-changed","enum-values-removed","removed-param","annotation-flip-to-destructive","param-mirrored-to-header","output-schema-added","output-schema-changed","deep-schema-undiffable","tool-added","tool-removed"],"postures":["monitor","guard","strict"],"default_posture":"guard","fail_mode":"fail_closed","credential_custody":"none - reuses the session the client already authenticated; reads only public tool contracts","install":{"package":"mcpindex-gate","config_wire":["claude_desktop","claude_code","cursor","gemini_cli","cline","zed","windsurf","vscode"],"sdk":["typescript_wrap","python_wrap"],"docs":"https://mcpindex.ai/docs"},"status":"tier0_deterministic_live_verified_end_to_end; tiers1to3_built_held_off_by_default_opt_in","tiers":[{"tier":"tier0_deterministic_contract_diff","state":"live"},{"tier":"tier1_cloud_corpus_lookup","state":"held_off_by_default_opt_in"},{"tier":"tier2_llm_consult","state":"held_off_by_default_opt_in"},{"tier":"tier3_behavioral_verifier","state":"held_off_by_default_opt_in"}],"honest_limits":["contract_diff_not_safety_verdict","tiers1to3_held_off_by_default_opt_in","default_build_egresses_nothing_fail_closed","behavioral_tier_clears_or_refutes_not_safety_oracle","calibrated_false_v1"]},"blast_radius":{"what":"labels the blast radius of each tool call before the agent acts: what it would do (read/write/delete/send), whether it can be undone, and whether it leaves the machine","method":"deterministic static classifier over the tool's declared contract (name, description, input/output schema); does not run the tool","fields":["action_type","side_effect_class","reversibility","egress","autonomy_ceiling"],"fail_mode":"fail_closed","ambiguity_posture":"grades_up_assumes_more_dangerous_class","privacy":"deny_by_construction - every field is a typed enum/hash/bool; no raw argument value is carried","default":"on_by_default_in_published_clients","clients":["@mcp-index/sdk (typescript)","mcpindex-gate (python)"],"honest_limits":["blast_radius_is_static_not_a_safety_verdict","reads_what_a_call_would_do_not_runtime_arguments","advisory_grade_orchestrator_decides"]},"endpoints":{"mcp":"https://mcpindex.ai/api/mcp","search":"https://mcpindex.ai/api/v1/search?q={query}","recommend":"https://mcpindex.ai/api/v1/recommend?task={natural_language}","preflight":"https://mcpindex.ai/api/v1/preflight?task={natural_language}","diff":"https://mcpindex.ai/api/v1/diff?since={YYYY-MM-DD}","driftAny":"https://mcpindex.ai/api/v1/drift/any?fp={tool_fingerprint}","driftLedger":"https://mcpindex.ai/api/v1/ledger","detail":"https://mcpindex.ai/server/{slug}","llmsTxt":"https://mcpindex.ai/llms.txt","llmsFullTxt":"https://mcpindex.ai/llms-full.txt","verdictTool":"https://mcpindex.ai/api/v1/trust/tool/{server_id}/{tool_name}","verdictServer":"https://mcpindex.ai/api/v1/trust/server/{server_id}"},"docs":"https://mcpindex.ai/docs","license":{"terms":"https://mcpindex.ai/terms","registryMetadata":{"covers":"server names, descriptions, versions, install details","source":"https://registry.modelcontextprotocol.io","rights":"upstream_public_data - mcpindex claims none and grants none"},"researchDatasets":{"covers":"published drift and screening datasets","source":"https://zenodo.org","spdx":"CC-BY-4.0","rights":"reuse with attribution, commercial permitted, not narrowed by these terms"},"judgmentLayer":{"covers":"screen verdicts, MCP Quality Score, drift ledger, source-liveness evidence","holder":"Bhartis LLC","rights":"noncommercial use with attribution","permitted":["query via documented API","cite","link"],"prohibited":["bulk extraction to reconstruct or resell the dataset"],"commercialLicense":"hello@mcpindex.ai"}},"mcpServer":{"package":"mcp-server-mcpindex","version":"0.4.0","registry":"npm","tools":["recommend_mcp_for_task","search_mcp_servers","get_install_command","compare_servers","check_tool_trust","assess_server"]},"integrations":[{"framework":"mastra","package":"@mcp-index/mastra","registry":"npm","surface":"beforeToolCall hook that calls check_tool_trust (advisory screen) before a tool runs; warn or enforce; fail-closed, no credentials"}],"rateLimit":{"anonymous":"60 req/min/IP","contact":"hello@mcpindex.ai for higher limits"},"affiliation":"unofficial - not affiliated with Anthropic"}