Validate a Canonical Knowledge Structure. Returns validation result and diagnostics. Optionally accepts 'session_id' to validate an existing session's current state instead of creating a new one. Optionally accepts 'extensions' to opt into additional, non-default validation rules for this call only (see 'extensions' parameter). Returns a 'session_id' that can be used with list_versions and revert_version to track and manage version history.

Serialize a Knowledge Structure into its canonical JSON representation. Optionally accepts 'session_id' to serialize the current state of an existing session.

Produce a human-readable explanation of a Knowledge Structure. Optionally accepts 'session_id' to explain the current state of an existing session.

Apply structural evolution operators to a Knowledge Structure. Returns a new 'session_id' and 'version_id'. The 'session_id' can be used with list_versions and revert_version.

Three-way merge of Knowledge Structures. Provide a common ancestor (base) and two independently evolved branches. Returns the merged structure or a list of conflicts if automatic resolution is impossible.

Fork a new session from an existing one. Use this to isolate an experiment, explore an alternative modeling approach, or try a risky edit without touching the parent session -- if the branch doesn't pan out, close_session it; if it does, merge_branch it back into the parent.

Session-aware three-way merge: merge a branch session's changes into a target session. The merge base is resolved automatically from the branch's recorded fork point (set by create_branch), so -- unlike merge_knowledge -- you never supply the base yourself. On success, commits the merged result as a new version of the target session. On conflict, returns a 'conflicts' list (object_id, target_diff, source_diff) instead of merging. Do not call merge_branch again unchanged after a conflict -- retry it with a 'resolutions' argument covering each conflicting object_id (see the 'resolutions' parameter), which merges everything -- non-conflicting changes and now-resolved conflicts alike -- in this one call; identities you don't supply a resolution for are reported again. Only if you'd rather change the target session's content directly, apply your resolution there with evolve_knowledge and retry merge_branch with no resolutions. Either way, close_session the source branch once it has been fully integrated.

Close a session, releasing it from the runtime. Typical use: after merge_branch reports success, close_session the source branch that was just merged in -- it has been integrated and no longer needs to stay open.

Extract the local k‑hop neighbourhood around one or more seed ids from a session's current Knowledge Structure. Returns a self‑contained subgraph (serialized) and metadata: total_found_nodes, returned_nodes, is_truncated, truncation_reason, suggested_next_seed. Use filters (include_relation_types, include_object_types) to narrow the traversal, and max_tokens/max_objects to cap the result. type_weights can prioritise certain object types when the budget forces truncation.

Semantically search the Knowledge Structure of a session. Provide a natural language query; if the storage backend has a vector index (embeddings generated via the background outbox worker), matching seed objects are found automatically. Pass explicit 'seed_ids' instead when you already know which objects to expand around, or as a fallback if no embeddings have been generated yet for this session. The tool expands the neighbourhood around the matched seeds using query_subgraph. Use this when you don't know exact object IDs but have a description of what you're looking for.

Return runtime metrics and the tool telemetry dashboard. 'runtime_metrics' contains invocation counts and average execution times per runtime operation type. 'tool_telemetry' contains per-MCP-tool call counts, success rates, latency percentiles (p50/p95/p99), and top error types since the server started.

Verify an external source by performing a real HTTP request. Creates a VerificationRecord that can be validated.

List all available versions of a session's history. Requires a 'session_id' obtained from a previous call to validate_knowledge or evolve_knowledge.

Revert a session's Knowledge Structure to a specific previous version. Requires a 'session_id' obtained from a previous call to validate_knowledge or evolve_knowledge.

Compare the current state of a session against a target version. The returned diff is directional. 'direction' explicitly describes how to interpret the changes. 'base_version_id' is the historical version being compared against. 'target_version_id' is the current session state. The response also contains a semantic summary (added/removed objects and relations) to make interpretation easier for LLMs.

Export a subgraph as a Mermaid diagram. Many MCP clients render Mermaid natively; if yours doesn't, the raw Mermaid text is still useful as structured output. Use this after query_subgraph to show the structure.

Explain the differences between the current state of a session and a target version in plain English. Useful for understanding what changed without parsing raw diff output.

Export a session's Knowledge Structure to another format. Supports 'json-ld', 'turtle', and 'rdf-xml'.

Given a session and a description of what to change, return the current objects/relations and guidance for constructing valid evolution operations. Use this before evolve_knowledge to reduce trial-and-error. If you already have a candidate 'operations' list (same format evolve_knowledge accepts), pass it here first to preview whether it would be valid -- this dry-runs it the same way evolve_knowledge does internally, but commits nothing, so you can check correctness before spending a real evolve_knowledge call on a guess.

Detect logical contradictions in a Knowledge Structure using the contradiction/conflict extension constraints. Supports three types of detection:

• mutual_exclusion: Flags when the same source-target pair has both of two declared relation types.

• functional_relation: Flags when a source has multiple targets via a declared single-valued relation type.

• inference_confidence_conflict (see ADR-001): Flags when two or more active (non-superseded) InferenceStep objects share a 'conclusion' but disagree on 'confidence'. Reported at WARNING severity, not ERROR -- this is a resolvable belief conflict between agreeing inference paths, not a jointly-nonsensical relation pair. Mark a step no longer active with its own 'superseded_by' field, not by editing another step. Examples of contradiction rules:

• MutualExclusionRule: {"identity": {"id": "rule-1", "type": "MutualExclusionRule", "name": "no-support-and-refute"}, "structure": {"relation_type_a": "supports", "relation_type_b": "refutes"}}. This flags when the SAME source-target pair has BOTH a 'supports' and a 'refutes' relation.

• FunctionalRelationRule: {"identity": {"id": "rule-2", "type": "FunctionalRelationRule", "name": "single-orbit"}, "structure": {"relation_type": "orbits"}}. This flags when a single source has MORE THAN ONE target via 'orbits'. To use mutual_exclusion/functional_relation, ensure your structure contains MutualExclusionRule and/or FunctionalRelationRule objects. inference_confidence_conflict needs no rule object -- it applies to any InferenceStep objects present.

Create an isolated sandbox branch from a parent session, optionally apply a hypothesis (list of evolution operations) immediately, and show how the sandbox differs from its fork point. The parent session is never touched. Safe to discard with close_session if the hypothesis doesn't pan out.

Build a Canonical Knowledge Structure from free-form text using an LLM. The LLM extracts entities and relationships, generates a valid CKS JSON payload, which is then parsed and validated before being persisted as a new session. Provider is auto-selected (CKS_LLM_PROVIDER): a local Ollama server if reachable (no API key needed), else Anthropic if ANTHROPIC_API_KEY is set. Returns 'session_id', 'version_id', and the serialized structure. Use 'hint' to direct the extraction toward specific aspects of the text.

Export a full session bundle for migration or archival. Unlike export_knowledge (which converts to RDF/JSON-LD), this tool packages the session's current structure, version history, and metadata into a self-contained JSON document that can be used to recreate the session in another runtime instance. Supports two formats: 'bundle' (default) — a complete migration envelope with version history; 'cks' — bare canonical CKS JSON of the current structure only. Set 'include_structures' to true to embed the full KnowledgeStructure for each historical version (may be large).

Fetch a public URL, extract its title, description and key topics, and return a Knowledge Structure representing the document. The document object is linked via 'mentions' relations to Topic objects for each extracted keyword. SSRF protection is applied, so private/internal URLs are refused.