GovernancePlatform MCP Server

Validate structured data against a Blueprint's rules. Returns PASS, FAIL, or REVIEW. The platform checks mathematical accuracy (do the numbers add up?), structural consistency (do the fields satisfy all constraints?), and semantic plausibility (do the values make sense in context?). Every result includes a determinism hash — the same input with the same Blueprint always produces the same result. Auditable, replayable, legally defensible. A Blueprint is required for meaningful validation. Without one, use create_blueprint or load_rule_pack to define your governance rules first. Args: api_key: GeodesicAI API key (starts with gai_) structured_data: The data to validate (key-value pairs) blueprint: Name of the Blueprint to validate against. Use list_blueprints to see options.

Validate structured data and automatically compute repairs if it fails. Single call that combines validate + repair. Different from validate: validate returns only the verdict; if the data fails, you'd then call repair separately. validate_repair returns the verdict AND the repaired payload in one call. Different from repair: repair always returns repair suggestions regardless of whether the input was valid; validate_repair only computes repairs when validation actually fails. If PASS: returns the validated data with determinism hash. If FAIL: returns the failure details AND a repaired payload with field-by-field corrections and confidence scores. The agent can inspect the repairs and resubmit the corrected data. If REVIEW: returns the flagged data with review reasoning. This is the recommended starting point for most agent integrations. Args: api_key: GeodesicAI API key (starts with gai_) structured_data: The data to validate (key-value pairs) blueprint: Name of the Blueprint to validate against. Use list_blueprints to see options.

Create a Blueprint — a governance contract that defines validation rules. A Blueprint tells the platform what "correct" means for your data: which fields exist, what math must hold between them, and what value ranges are acceptable. Without a Blueprint, the platform has nothing to validate against. If you don't know what rules to define, use load_rule_pack to start from a prebuilt template, or use discover_patterns to find rules from your data. Use the blueprint_guide prompt for the complete reference of all available rule types, constraint types, and configuration options. Args: api_key: GeodesicAI API key (starts with gai_) customer_name: Organization or project name (used for folder naming) workflow_name: Unique identifier for this Blueprint (used as the 'blueprint' parameter in validate) mode: "observe" (platform checks agent's work) or "enforce" (platform computes derived fields) extracted_fields: Fields the agent extracts from source data (e.g. ["vendor", "qty", "unit_cost"]) derived_fields: Fields computed from other fields (e.g. ["subtotal", "total"]) derivation_rules: Math rules defining field relationships. Available types: "add" (target = a + b), "subtract" (target = a - b), "multiply" (target = a × b), "divide" (target = a ÷ b), "round" (round field to N places), "copy" (copy source to target), "items_multiply" (per-item a × b in a list), "items_sum" (sum a field across list items). Each rule requires "type" and the relevant fields. See blueprint_guide prompt for full schema. formal_constraints: Value bounds and ratio constraints. Available types: "magnitude_anchor" (field within min/max range, requires "field", "min", "max"), "relative_anchor" (ratio a/b within tolerance, requires "a", "b", "expected_ratio", "tolerance"), "max_action_threshold" (trigger action if field exceeds threshold, requires "field", "threshold", "action"). See blueprint_guide prompt for full schema. semantic_checks: Domain-specific validation checks require_math: Validate mathematical relationships (default true) require_consistency: Check internal consistency (default true) require_coherence: Check structural coherence (default true) require_provenance: Require agents to report extraction source locations require_high_assurance: Strictest validation — feasibility, spectral, and global consistency required enable_anomaly_detection: Geometric fingerprinting to detect structural outliers enable_drift_tracking: Monitor pattern stability across batches

List Blueprints owned by the calling account. Returns each Blueprint's name, workflow identifier, mode, and field/rule/constraint counts. Use the workflow_name as the 'blueprint' parameter when calling validate. Blueprint modes — important for agents that summarize results: - "observe" (Observation mode): the platform validates data against the Blueprint's rules and returns PASS, FAIL, or REVIEW with repair suggestions. Validation is detection-only; the caller decides what to do with the result. - "enforce" (Execution mode): the platform performs the same validation AND authorizes downstream side effects. Side effects are blocked if validation fails. In this mode, rules that compute derived values (like totals) publish the platform's computed result as authoritative, so downstream consumers see the canonical value rather than the agent's. Both modes run the same rule checks. The difference is what happens after validation, not during it. Scope: only Blueprints created by the calling account are returned. Use validate / repair / etc. with the workflow_name to operate on them. Args: api_key: GeodesicAI API key (starts with gai_)

Compute the nearest valid point on the constraint manifold. Given structured data with errors, missing values, or inconsistencies, returns field-by-field repair suggestions with geometric confidence scores. Uses algebraic projection for linear constraints and iterative geodesic projection for nonlinear constraints. Can accept rules directly or load them from a Blueprint. Args: api_key: GeodesicAI API key (starts with gai_) structured_data: The data to repair (key-value pairs) derivation_rules: Math rules (optional if blueprint is provided) formal_constraints: Constraints (optional if blueprint is provided) blueprint: Load rules from this Blueprint instead of specifying them directly

Check whether a set of constraints can be simultaneously satisfied. Uses the twist-compression operator to detect structural obstructions — fundamental conflicts in the constraint system that cannot be resolved by adjusting any single field. Returns obstruction type, magnitude, and the specific constraint interactions causing the conflict. Fast pre-check before running full validation. Args: api_key: GeodesicAI API key (starts with gai_) constraints: List of derivation rules and formal constraints to check field_values: Current field values (numeric key-value pairs)

Deterministic forward reasoning — show what valid states are reachable. Given the current data state, generates candidate next states by applying Blueprint rules, then ranks them by structural quality (lowest drift, highest confidence, shortest path, or lowest risk). Args: api_key: GeodesicAI API key (starts with gai_) structured_data: Current data state blueprint: Blueprint to use for rules and constraints max_depth: How many steps ahead to forecast (1-10) max_branches: Maximum branches per step (1-10) rank_by: Ranking criterion — "drift" (most stable), "confidence" (highest geometric confidence), "shortest" (fewest changes), "risk" (lowest risk)

Feed a batch of structured data to discover patterns deterministically. No Blueprint required — discovers rules from the data itself. Analyzes the batch using motif discovery, structural routing, and geometric fingerprinting. Returns candidate validation rules with confidence scores, structural type classifications, and anomaly counts. Source data is never stored — only statistical summaries persist. Discovered rules can be promoted into Blueprints using approve_rule. Args: api_key: GeodesicAI API key (starts with gai_) documents: List of structured data objects to analyze blueprint: Namespace for this discovery session

Find the shortest path from an invalid state to a valid one. Given data that fails validation, computes a sequence of minimal field changes that would bring the data into compliance with the Blueprint's rules and constraints. Returns the path as an ordered list of intermediate states. Different from repair: repair returns the single nearest valid point on the constraint manifold (a one-shot projection). repair_path returns the trajectory between invalid and valid states — useful when you want to see what's changing step by step, or when there are multiple ways to repair the data and you want to evaluate alternatives. Use repair when you just need the fix; use repair_path when you need to explain or audit the fix. Args: api_key: GeodesicAI API key (starts with gai_) structured_data: Current (invalid) data state blueprint: Blueprint defining the valid constraint space. Caller must own the Blueprint. max_depth: Maximum repair steps to search (1-10). Higher values find paths through more complex repairs but take longer. rank_by: Ranking criterion for selecting among valid paths. One of: "shortest" — fewest changes "drift" — most stable trajectory "confidence" — highest geometric confidence "risk" — lowest risk Returns: path: ordered list of intermediate states, each with field values and structural quality metrics path_length: number of steps in the returned path terminal_state: the final valid state at the end of the path rank_score: score of the chosen path under rank_by alternatives: summary of paths considered but not chosen

Compare outcomes under different rule sets. Given the same data, runs trajectory analysis under two different sets of rules/constraints and shows how the valid state space differs. Useful for what-if analysis: "what happens if I change this rule?" Args: api_key: GeodesicAI API key (starts with gai_) structured_data: Data to analyze blueprint: Primary Blueprint (rule set A) rules_b: Alternative derivation rules (rule set B) constraints_b: Alternative constraints (rule set B)