Oxyo MCP Server (oxyo-mcp)

Load a compact session-start context capsule (<350 tokens) returning active contradiction-free project/global beliefs, recent actions, root-cause warnings, and verified skills. Call once at cold start instead of `oxyo_recall`; use `oxyo_recall` for targeted mid-session searches.

Search persistent memory by keyword and meaning, returning compact episodic snippets, active contradiction-free beliefs, root-cause chains, and verified Python skills. Use for targeted mid-session lookups; call `oxyo_session_pack` instead at session cold start.

Append a timestamped task action, architectural decision, or observation to episodic history, returning the stored episode metadata. Use for chronological event logs; use `oxyo_remember_belief` instead for standing rules or facts.

Persist a durable fact, project rule, or user preference, automatically deduplicating paraphrases and superseding contradicted older 1:1 claims while preserving 1:N additive facts. Use for standing truths; use `oxyo_remember_episode` for transient chronological events.

Retrieve the immutable audit log of superseded beliefs paired with the newer claims that replaced them. Use to inspect historical rule changes and contradiction resolutions; use `oxyo_recall` for active beliefs.

Record or reinforce a directed cause-and-effect link in the project root-cause graph, returning updated edge confidence and observation counts. Use when logging newly verified bug causes or system dependencies; use `oxyo_causal_infer` to query paths.

Query the root-cause graph to trace upstream failure sources (`diagnose`), downstream blast radius (`intervene`), simulated interventions (`do_calculus`, `counterfactual`), or highest-value diagnostic checks (`experiments`). Use to analyze recorded dependencies; call `oxyo_causal_record` to add new edges.

Execute deterministic Python math, financial `Decimal`, or algorithmic checks inside an isolated zero-network sandbox (`RESULT = ...`), returning exact `RESULT`, stdout, and optional boundary-fuzz results. Use for procedural Python verification; use `oxyo_prove_constraints` for declarative constraint solving.

Validate and persist a reusable Python function into the skill library only after 100% of supplied `test_cases` pass in isolation, returning verification status. Use to save new tested utilities; use `oxyo_skill_invoke` to list or execute existing skills.

List registered Python skills (`action='list'`) or execute a named skill (`action='execute'`) with arguments in isolation, returning its deterministic return value. Use to run already-verified skills; call `oxyo_skill_register` first to create a new skill.

Compact older episodic history by archiving redundant entries and promoting recurring patterns while preserving the latest episodes, returning compaction statistics. Use for fast episodic cleanup; use `oxyo_sleep_abstract` when also distilling code traces into skills.

Manage persistent multi-session task dependency graphs (`create`, `decompose`, `next`, `update`, `tree`), returning unblocked next tasks and automatically propagating failure blockers. Use for multi-step project roadmaps across sessions; use `oxyo_remember_episode` for single-step logs.

Distill verified code execution traces into tested, reusable functions in the persistent skill vault while compacting older episodic logs. Use after coding milestones to extract skills; use `oxyo_consolidate` for episode cleanup without skill synthesis.

Solve bounded discrete constraint systems (`action='satisfy'` with optional `minimize`/`maximize`) or prove universal invariants (`action='prove'`, returning a formal proof or minimal counterexample). Use for declarative constraint/invariant models; use `oxyo_verify_python` for general Python scripts.

Generate and deterministically verify synthetic parametric coding challenges in isolation, returning curated verified solution traces and optionally promoting reusable helper skills. Use for autonomous skill warm-up and trace curation; use `oxyo_verify_python` to check a single snippet.