convez.ai
MCP ServerSegurançaIntermediário
Mipiti

Mipiti MCP Server

Mipiti/mipiti-mcp

Threat modeling e compliance automation para agentes IA

Ver no GitHub
InicialPluginScore · de 0 a 100

Servidor MCP que conecta agentes de IA de coding (Claude, Cursor) à plataforma Mipiti de postura de segurança. Permite gerar e gerenciar modelos de ameaça, controles, conformidade e evidências programaticamente através de 133 ferramentas especializadas.

Instalação

$npx convez-ai Mipiti/mipiti-mcp

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
1
Forks
0
Acompanhando
0
Saúde
62%
Último commit
há 1 dia
Versão
v0.106.1
Issues abertas
0
Contribuidores
3

Casos de uso

  • Gerar modelos de ameaça automatizados a partir de descrições de features
  • Mapear conformidade e controles de segurança programaticamente
  • Documentar postura de segurança com agents de IA
  • Integrar análise de risco em pipelines de desenvolvimento

Veredito

Ferramenta sólida de threat modeling para times que combinam segurança com automação via IA. Deploy fácil (OAuth ou API key), 133 tools especializadas, mas requer familiaridade com modelos de ameaça para máxima utilidade. Bom para acelerar compliance e documentação de segurança em larga escala.

README

A tradução para o português deste README ainda está na fila.

Mipiti MCP Server

MCP (Model Context Protocol) server for Mipiti — security posture platform.

Lets AI coding agents (Claude Code, Claude Desktop, Cursor, etc.) generate and manage threat models, controls, assumptions, compliance mapping, and evidence programmatically.

Hosted Endpoint (Recommended)

The Mipiti backend hosts an MCP server at https://api.mipiti.io/mcp. No installation needed — just configure your MCP client to connect.

Claude Code (quickstart)

claude mcp add --transport http Mipiti https://api.mipiti.io/mcp

You'll be prompted to log in via your browser (OAuth). That's it.

OAuth (manual config)

MCP clients with OAuth support (Claude Code, Claude Desktop, Cursor) automatically prompt you to log in via your browser. Add to your project's .mcp.json:

{
  "mcpServers": {
    "mipiti": {
      "type": "http",
      "url": "https://api.mipiti.io/mcp"
    }
  }
}

On first connection, your MCP client opens a browser window where you approve access with your Mipiti account. Tokens refresh automatically.

API Key

For clients without OAuth support, or headless/CI environments, create an API key in Settings:

{
  "mcpServers": {
    "mipiti": {
      "type": "http",
      "url": "https://api.mipiti.io/mcp",
      "headers": {
        "X-API-Key": "your-api-key"
      }
    }
  }
}

Standalone Package (Alternative)

If you prefer running the MCP server locally (e.g., for development or self-hosted instances), install the mipiti-mcp package. This is a thin HTTP client that calls the Mipiti API.

pip install mipiti-mcp
# Or run directly with uvx
uvx mipiti-mcp

Environment Variables

VariableRequiredDefaultDescription
MIPITI_API_KEYYes—Your Mipiti API key
MIPITI_API_URLNohttps://api.mipiti.ioAPI base URL
SERVER_VERSIONYes—Identifier for the running server's MCP surface (instructions, tool docstrings, schemas, behavior). Sent on every tool call. Clients invalidate cached MCP guidance when this changes. For local runs, any sentinel string is fine ("local", "dev"). For deployed runs, use a value that changes when this package's source changes (commit SHA is typical).

Claude Code (standalone)

{
  "mcpServers": {
    "mipiti": {
      "command": "uvx",
      "args": ["mipiti-mcp"],
      "env": {
        "MIPITI_API_KEY": "your-api-key",
        "SERVER_VERSION": "local"
      }
    }
  }
}

Tools (133)

Threat Modeling

ToolDescription
generate_threat_modelGenerate a complete threat model from a feature description. Runs a multi-step AI pipeline producing trust boundaries, assets, attackers, control objectives, and assumptions. Progress reported automatically via MCP protocol — the tool blocks until complete.
refine_threat_modelRefine an existing threat model based on an instruction. Creates a new version. Only affected entity types are modified — unaffected entities are preserved server-side.
query_threat_modelAsk a question about an existing threat model.
get_threat_modelGet the full details of a specific threat model (trust boundaries, assets, attackers, assumptions). Use include_cos=True to include control objectives.
list_threat_modelsList all saved threat models with IDs, titles, versions, and creation dates. Supports source filter and include_assessment_summary=True to inline per-model posture counts in one call (avoids N+1 looping assess_model).
rename_threat_modelRename a model (metadata only, no new version). Titles must be unique within a workspace (case-insensitive).
delete_threat_modelPermanently delete a model and all its data.
export_threat_modelExport as PDF, HTML, or CSV.
export_threat_model_archiveExport the self-contained JSON audit archive (every version, controls, assertions with CI verdicts, findings, attestations, sufficiency signatures). Independently verifiable.
import_threat_model_archiveRestore an audit archive into a target workspace. Fresh model_id per import; title collisions auto-suffix.

Entity CRUD

ToolDescription
add_asset / edit_asset / remove_assetTargeted single-entity changes for assets. Creates a new version.
add_attacker / edit_attacker / remove_attackerSame for attackers.

Trust Boundaries

ToolDescription
get_threat_modelReturns existing trust boundaries (along with assets, attackers, assumptions). Review current boundaries before adding or modifying.
add_trust_boundary / edit_trust_boundary / remove_trust_boundaryCRUD for trust boundaries. Defines where trust transitions occur in the system architecture. Attackers are positioned at boundaries; COs are annotated with boundary reachability. Changes auto-generate boundary assumptions for newly unreachable COs.

Controls

ToolDescription
get_controlsList controls with current status. Use summary_only=True for compact response.
get_control_objectivesList COs with which controls cover each one. Pair with get_reachability_verdicts for per-CO composer reachability state.
update_control_statusMark implemented or not_implemented. Requires at least one assertion first.
refine_controlModify a control's description with justification. Platform evaluates whether the mitigation group still covers the COs.
regenerate_controlsRegenerate controls. Supports mode="per_co" and co_ids to target specific COs.
import_controlsImport controls from JSON or free text, auto-mapped to COs and deduplicated.
delete_controlSoft-delete with justification. Blocked if it's the only control covering a CO.
check_control_gapsAI-powered gap analysis across all controls.
get_mitigation_groups / set_mitigation_groupsInspect and modify how controls are grouped into mitigation paths for a CO (AND within groups, OR across groups). Platform AI-evaluates whether proposed changes preserve CO coverage.
set_co_calSet per-CO ISO/SAE 21434 Cybersecurity Assurance Level (1-4). Persisted on the control_objectives identity side-table; survives soft-delete + revival; no new model version.

Assumptions and Attestation

ToolDescription
get_threat_modelReturns existing assumptions (along with assets, attackers, trust boundaries). Review current assumptions before adding or modifying.
add_assumptionAdd an assumption, optionally linking it to COs via linked_co_ids.
edit_assumptionUpdate description and/or linked COs.
remove_assumptionSoft-delete (preserved for audit). Linked COs are no longer mitigated by it.
restore_assumptionRestore a soft-deleted assumption. Re-attestation required.
submit_attestationRecord that a responsible party affirmed an assumption holds. Provide attested_by, statement, expires_at.
list_attestationsAttestation history for an assumption.
assume_controlShorthand: mark a control as externally handled by a single assumption (writes to group 1). Counts as active for mitigation group completeness when attested.
unassume_controlShorthand: clear externally-handled status; control reverts to not_implemented. Removes all assumption groups.
get_control_assumption_groupsInspect the current assumption group structure on a control. Groups express alternative sets of external claims (within = AND, across = OR).
set_control_assumption_groupsDeclaratively set the assumption group structure. Use for compound cases (e.g. "AWS KMS + quarterly review") or multiple independent paths.
convert_assumption_to_controlsGenerate controls for assumption-covered COs and retire the assumption linkage.

Assertions and Evidence

ToolDescription
submit_assertionsSubmit typed, machine-verifiable claims about system properties (28 assertion types).
list_assertions / delete_assertionList or delete assertions for a control.
add_evidence / remove_evidenceAttach auxiliary metadata (docs, links). Evidence is contextual — only assertions prove implementation.
get_verification_reportShows verified, partially verified, and unverified controls with sufficiency details.
get_sufficiencyQuick check: do assertions for a single control collectively cover all aspects?
get_scan_promptReturns targeted prompts for scanning the codebase against not_implemented controls.
get_review_queueControls not reviewed in 90+ days. Start here for periodic maintenance.
submit_findings / list_findings / update_findingReport and track negative findings (gap discovery).
preview_finding_remediationRead-only. Returns a structured diff describing the changes a subsequent apply_finding_remediation call would make. Diff shape depends on the finding's kind (e.g. for structural_duplicate_controls: which controls would be kept, which dropped, the union of CO mappings + framework refs that would land on the survivor). Call before apply_finding_remediation so the operator can confirm.
apply_finding_remediationMutates state: commits the changes preview_finding_remediation showed. Requires a non-empty justification (one-line operator rationale) recorded on the audit trail. The agent is responsible for the preview-then-apply norm — surface the diff and get explicit confirmation before calling.

Assurance

ToolDescription
assess_modelDeterministic assessment of all COs. Returns mitigated/at_risk/unassessed with risk_reason (missing_controls, pending_attestation, expired_attestation). For per-CO reachability state call get_reachability_verdicts.
get_findings_risksWorkspace-scoped triage dashboard: open findings, active risk acceptances, and at-risk COs across every model the workspace can access. Entry point when asked "what's open?".
get_model_risk_viewPer-model Prioritized Risk View: one row per live CO with derived risk tier, asset impact, attacker likelihood, control coverage, and open-finding count.
get_system_risk_viewCross-model variant of get_model_risk_view: same shape, aggregated across every model in a System (model_id + model_title attached per row).
list_risk_acceptancesAll risk acceptances on a model — risks explicitly accepted instead of mitigated. Includes CO id, owner, justification, status, review deadline.

Composition (recursive-tree effective model)

Views over the effective model — own entities composed with everything inherited from ancestor threat models on the recursive tree. Backend-gated by TREE_COMPOSITION_ENABLED; when off, read tools return a stable empty body with flag_enabled: false and the write tool returns 503.

ToolDescription
get_composition_overviewIndex: counts + tree metadata (parent_id, ancestor_chain, depth, child_ids) + structural warnings. Cheapest call — use first to learn whether composition is enabled and orient on the tree.
list_effective_entitiesEffective entity set keyed by kind (trust boundaries, components, assets, attackers, attack paths). Each entry carries provenance (own vs inherited) and a fully-qualified id for cross-model references.
list_effective_control_objectivesEffective COs tagged with origin (own / cross / inherited). Pair with get_effective_coverage and get_reach_verdicts.
get_effective_coveragePer-CO coverage with credited inheritance: own_credit, inherited_credit, and the list of contributing controls (with the owning model id, origin, verification status, mitigation group). This is what drives the composition coverage view, not per-model get_verification_report.
get_reach_verdictsPer-CO reachability verdicts over the composed effective topology — same kinds (reachable / unreachable / indeterminate) as get_reachability_verdicts, but evaluated against the merged tree. Use on child models when ancestor topology matters.
list_effective_attack_pathsEffective AttackPath set + lifted missing/dangling suggestions computed against the composed reach surface.
list_reconciliation_candidatesPaginated reconciliation candidates between this model and its ancestors. Tier certain is a deterministic match safe to auto-apply; tier heuristic is fuzzy and needs review.
apply_certain_reconciliation_matchMutating. Apply a certain-tier candidate from list_reconciliation_candidates: soft-deletes the descendant's own duplicate so the inherited entity becomes canonical. Server re-validates against current live state and refuses heuristic-tier candidates (those need operator-driven structural-divergence review). Bumps model version; returns the standard _do_entity_crud envelope ({model, controls_carried, controls_orphaned, orphaned_control_ids}).
reject_reconciliation_candidateMutating. Persist the operator's "these are NOT duplicates" decision at org scope so the candidate detector filters this pair out of the active queue on subsequent reads. Idempotent on the natural key (model_id, kind, own_qid, inherited_qid). Does NOT bump model version (rejection is org state). Returns the persisted record — keep the id if the operator may unreject later.
unreject_reconciliation_candidateMutating. Remove a persisted rejection by surrogate id (from list_reconciliation_rejections or the return value of reject_reconciliation_candidate). The pair becomes eligible to surface in the active queue again on the next read. Returns {ok: true}.
list_reconciliation_rejectionsList the persisted rejections on a model in rejected_at ascending order — the same set the candidate detector consults to filter the active queue. Use to render a rejected section in a triage view or to find the surrogate id needed by unreject_reconciliation_candidate.
lift_composition_entityMutating. Promote a shared-anchor entity from two sibling descendants to their lowest common ancestor: each source's copy is soft-deleted and the inherited entity becomes canonical for every descendant of the LCA. Server re-detects field-level and attached-state conflicts against current live state; pass field_resolutions / attached_state_resolutions keyed by the conflict keys returned in the 400 detail. Server also runs an over-application gate against the LCA's descendant set; pass acknowledged_third_party_subtrees to acknowledge extra reach or skip_overapplication_gate=true to override after explicit operator confirmation. Bumps version on the LCA + both source descendants; returns {lift_id, lca_model, descendant_a_model, descendant_b_model, applied_migrations, lift_event} — the lift_event block matches the audit pack's lift_history entry.
split_composition_entityMutating. Inverse of lift_composition_entity: push an ancestor-owned entity down to one or more target descendants and soft-delete the ancestor's copy. A new local id is minted on each target; attached state (assertions, jira mappings, risk acceptances) on the ancestor's entity is duplicated to every target. Bumps version on the ancestor + every target descendant; returns {split_id, ancestor_model, descendant_models, applied_duplications, split_event} — the split_event block matches the audit pack's split_history entry.
preview_undo_lift_compositionRead-only. Compute the inverse plan (or divergence refusal) for a prior lift_applied event WITHOUT mutating state. Returns {plan, refusal} — exactly one is non-null. Surface this to the operator before calling undo_lift_composition_event so the confirmation step shows the state operations the apply would commit, or the enumerated reasons the divergence detector would refuse.
undo_lift_composition_eventMutating. Apply the inverse of a previous lift_applied event. Re-runs the divergence detector immediately before applying and refuses with 409 + a structured refusal block (detail.refusal.reasons) when state has materially evolved since the forward lift. On success, persists the inverse state operations across the LCA + every affected source descendant and emits a structured lift_undone activity event citing original_event_id so the audit pack chains undo to its forward. Returns {undone_event_id, original_event_id, applied_state_ops, models: {lca_model, source_descendant_models}}.
preview_undo_split_compositionRead-only. Counterpart to preview_undo_lift_composition for splits. Same {plan, refusal} shape; the plan block carries the split-specific inverse operations (restore at the ancestor, tombstone the duplicated copies on every target descendant).
undo_split_composition_eventMutating. Mirror of undo_lift_composition_event for splits. Same divergence-detector contract — refuses with 409 + structured refusal block when state has evolved. On success, restores the ancestor's entity, tombstones the duplicated copies on every target descendant, and emits a structured split_undone activity event citing original_event_id. Returns {undone_event_id, original_event_id, applied_state_ops, models: {ancestor_model, descendant_models}}.

Cross-model dependencies (delegation)

Declared reliance edges (distinct from the parent/composition tree, which is containment): a model depends on a control implemented in another model — for systems built on shared services (auth, logging, shared data) rather than sub-parts. The target is always a provider control (credit terminates at a proven mechanism). Reliance is workspace-scoped: a consumer can only delegate to provider models in the same workspace (these tools don't see models across workspace boundaries). Backend-gated by RECURSIVE_TREE_ENABLED; credit effects further gated by FOUNDATION_DELEGATION_ENABLED.

ToolDescription
declare_foundationMark a shared-service model as a foundation that advertises specific controls (provides) other models can delegate to. A capability advertises a control, never an objective.
create_relianceDeclare a single dependency. delegated (consumer has no local control for an objective; provider handles it — pass source_objective_id) or relied_upon (consumer keeps its own control but its validity depends on the provider's — pass source_control_id). Enters draft; runs LLM semantic validation.
confirm_reliancePromote a draft edge to active — the credit-soundness gate. Refused unless validation returned valid; a partial/mode-mismatch is never silently credited.
delete_relianceRemove a reliance edge.
list_relianceA model's dependency edges (as consumer) plus who relies on it (as provider — the blast radius before changing its controls).
propose_attach_foundationRead-only. Propose which of a consumer's objectives each foundation capability covers (scored). Feed the chosen subset to attach_foundation.
attach_foundationBulk-create draft delegation edges for the selected (objective, provider control) pairs. Each runs LLM validation; none credits until confirmed.

Tags (grouping)

Overlapping, semantics-free grouping of models (the Affiliation primitive) — for audit scopes, ad-hoc selections, or portfolios. A model may carry many tags; a tag never affects posture or credit.

ToolDescription
create_tag / delete_tagCreate or remove a tag (deleting affects the grouping only, not the member models).
add_model_to_tag / remove_model_from_tagManage membership; a model can belong to many tags at once.
list_tags / list_model_tagsBrowse the workspace's tags, or a model's tags.
get_tag_risk_viewAggregate per-CO risk across a tag's members. Delegation-aware (a CO mitigated via a verified cross-model delegation reads as covered).
select_tag_compliance_frameworksMake a tag a compliance/audit scope: select frameworks for the tag, propagated to its members.
get_tag_compliance_reportCross-model compliance coverage report scoped to a tag's members (the tag equivalent of the system compliance report).
export_tag_reportSigned auditor HTML for a tag — member reports + cross-model dependency graph + attestation status (the tag equivalent of the system auditor export).

Compliance

ToolDescription
list_compliance_frameworksAvailable frameworks (OWASP ASVS, ISO 27001, SOC 2, NIST CSF, GDPR, FedRAMP, PCI DSS, EU CRA).
select_compliance_frameworksSelect frameworks for a model.
get_compliance_reportCoverage report for a selected framework.
auto_map_controlsAI-powered semantic mapping of controls to framework requirements.
map_control_to_requirementManual control-to-requirement mapping.
auto_remediateLLM-powered gap closure — proposes new assets, attackers, and controls for uncovered framework requirements.

Components

ToolDescription
add_component / edit_component / remove_componentComponents bridge trust boundaries (security architecture) to repositories (code organization). Component(id, name, repo_url, path, trust_boundary_ids) scopes controls to the codebase that implements them. Used for multi-repo systems and per-repo threat models. edit_component also accepts optional per-component level grades: target_sl (IEC 62443 Security Level, 1-4), eal (Common Criteria Evaluation Assurance Level, 1-7), fips_level (FIPS 140-3 Security Level, 1-4).

Systems and Workspaces

ToolDescription
list_workspacesList available workspaces.
update_organizationSet per-organization level grades: target_ml (IEC 62443-4-1 Maturity Level, 1-5), csf_tier (NIST CSF Tier, 1-4). Admin-only. Use clear_target_ml / clear_csf_tier to explicitly reset to NULL.
list_systems / get_system / create_systemManage systems (groups of related models).
add_model_to_systemAdd a model to a system.
get_system_dependenciesCross-model dependency graph with satisfaction status for assumptions linked to other models.
link_dependencyLink a cross-model assumption to a target model — dual-path satisfaction (controls OR manual attestation).
select_system_compliance_frameworks / get_system_compliance_reportSystem-level compliance aggregation.

Setup and Operations

ToolDescription
get_setup_statusCheck which onboarding steps are done.
complete_setup_stepMark an onboarding step as done (mcp_configured, mipiti_verify_installed, ci_secret_added, ci_pipeline_added).

Development

git clone https://github.com/Mipiti/mipiti-mcp.git
cd mipiti-mcp
pip install -e ".[dev]"
python -m pytest -v

Local Testing with Claude Desktop

{
  "mcpServers": {
    "mipiti": {
      "command": "uv",
      "args": ["run", "--directory", "/path/to/mipiti-mcp", "mipiti-mcp"],
      "env": {
        "MIPITI_API_KEY": "your-key"
      }
    }
  }
}

License

Proprietary. Copyright (c) 2026 Mipiti, LLC. All rights reserved. See LICENSE for details.

Ver o original no GitHub

Bom para

Relacionados