World Model MCP

Install

Works with Claude Code, Codex, VS Code, Cursor, Windsurf, Claude Desktop. Read the repository before you give any tool access to your data.

npx agentmarket install connector/putervision-world-model-mcp

About

Use cases: agent-memory

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this server can do and what it asks for, from its tool list, install config and code. Checked Oct 25, 2026.

Tools

6 read · 8 write · 1 delete

Secrets

None asked for

Install scripts

None found

Runs as

Node.js package

Tools (15)

Read from the source code; the running server may differ. Descriptions are the server’s own words; hidden characters are shown as [U+…].

• create_evidence_pack

  Package entity positions, observation reconciliations, and snapshot states into an immutable, SHA-256 hashed cryptographic evidence pack for compliance and state-memory task verification. Does not change entities; hashes current proof. Returns {ok, pack_id, hash, payload}. Use create_evidence_pack instead of manage_snapshot when creating immutable cryptographic verification packages rather than database checkpoints.

• generate_game_inputs

  Translate 3D navigation paths into Playwright commands (WASD/click-to-move) or project/unproject 3D coordinates and screen pixels. Actions: generate_inputs, project_screen, unproject_ray. Read-only: generates Playwright inputs, does not press keys. Returns {ok, action, inputs[]|screen_coords|ray}. Use generate_game_inputs instead of simulate_movement when translating 3D trajectories into Playwright browser automation commands.

• get_expected_view

  Compute which entities should be visible from an observer pose and FOV cone, with ray-AABB occlusion. Read-only. Does not write entities. Returns {visible[], occluded[], observer}. Use get_expected_view instead of get_spatial_map when computing observer FOV visibility and occlusion cones rather than unfiltered world states.

• get_spatial_map

  Export the known world as json, geojson, topological_graph, gltf, obj, joint, spatial_vlm, summary, or compact_slice. Optional region_id and min_confidence filters. Read-only snapshot of current SQLite state (not a live renderer). Returns the payload in requested format. Use get_spatial_map instead of query_entities when exporting full environment snapshots or 3D meshes rather than filtering entities.

• ingest_observation

  Merge vision detections into the world model: re-identify by Euclidean proximity, boost confidence, and optionally reconcile against the expected frustum. This is the perception writer. It may create or update entities when reconcile=true. Returns {ok, matched[], created[], lost[], reconcile}. Use ingest_observation instead of update_entity when merging camera or sensor perception detections rather than manual authoring.

• link_to_goal

  Link or unlink entities and regions to state-memory task nodes, or extract a goal-relevant spatial slice. Actions: link, unlink, get_context. link/unlink mutate association rows only. get_context is read-only. Returns {ok, action, links[]|context}. Use link_to_goal instead of record_outcome when associating entities with state-memory task nodes rather than recording physical movement deltas.

• manage_snapshot

  Checkpoint, diff, undo, or time-travel the spatial world database. Actions: save, restore, diff, list, undo, history, time_travel. restore (overwrites live state) and undo (reverts last matching mutation) are destructive and not always reversible except by saving first. diff, list, history, and time_travel are read-only. Returns {ok, action, snapshots[]|diff|state}. Use manage_snapshot instead of update_entity when rolling back, diffing, or time-traveling database state rather than editing single entities.

• manage_spatial_spec

  Manage Spatial Spec-Driven Development (Spatial SDD) physical baseline contracts. Actions: set (registers a baseline spec), verify (evaluates live entities against constraints without mutating), list (returns all registered specs). verify is read-only. set persists constraints. Returns {ok, action, spec_id, passed:boolean, violations[]}. Use manage_spatial_spec instead of update_entity when validating physical contract baselines rather than mutating live entity state.

• query_entities

  Find entities by keyword query (FTS5 search), type, region, spatial proximity, tags, or status. Alternatively, provide entity_id for single-entity location and historical trajectory lookup. Read-only. Returns {ok, count, entities[]}. Use query_entities instead of get_spatial_map when searching for specific subsets rather than exporting the full topology.

• record_outcome

  Record action execution results, movement deltas, property changes, entity destruction, or spool outcome ingestion from 60Hz loop. Not for vision ingest or manual pose edits. Mark entity destroyed via status: "destroyed". Returns {ok, action, success:boolean, ...}. Use record_outcome instead of update_entity when applying executed action results and status deltas rather than hand-authoring entities.

• set_relation

  Record or remove a spatial relationship between two entities. Actions: add, remove. Relations: on, inside, next_to, above, below, near, contains, occluded_by, connected_to, facing, holding, part_of, custom. This mutates the relation graph only, not entity poses. Returns {ok, relation}. Use set_relation instead of update_entity when establishing topological links (on, inside, contains) rather than setting entity coordinates.

• simulate_movement

  Predict entity trajectory, test for AABB obstacle collisions, or compute navigation waypoints (modes: simulate, navigate, waypoints). Read-only simulation. Returns {ok, is_valid, destination, collisions[], waypoints[]}. Use simulate_movement instead of record_outcome when testing hypothetical motion and collisions before executing an action.

• update_entity

  Create or upsert one spatial entity (position, orientation, AABB, tags, properties, confidence). Manual authoring only. Omit id to create (ULID assigned). Provide id to update. Omitted fields are preserved; this is a partial merge, not a full replace. status defaults to active. confidence is 0.0–1.0 object-permanence. Does not ingest vision detections or apply action results. Returns {ok, entity_id, created:boolean, entity}. Use update_entity instead of ingest_observation when authoring entities directly rather than merging perception detections.

• use_spatial_blackboard

  Publish or read multi-agent spatial coordination topics, collision alerts, and mutex region leases. Actions: get, set, delete, lease, list, post, read, claim, release. set writes payload (coordinates allowed for collision alerts) with optional ttl_seconds. lease acquire fails if the resource is held. delete is destructive. Returns {ok, action, items[]|entry|lease}. Use use_spatial_blackboard instead of set_relation when coordinating transient multi-agent collision alerts and mutex region leases.

• wait_for_spatial_state

  Poll and wait until an entity reaches a specific spatial condition (exists, active, confidence threshold, or enters region). Read-only polling tool. Default timeout: 10000ms, poll_interval: 500ms. On timeout returns {ok: false, timeout: true}. Returns {ok, condition_met:boolean, entity}. Use wait_for_spatial_state instead of query_entities when polling asynchronously for an entity state transition or region entry.

Signals

Stars

20

Forks

2

Open issues

1

Contributors

0

Last commit

Oct 3, 2026 (25 days ago)

Last release

Oct 3, 2026 (25 days ago)

From GitHub (fetched Oct 24, 2026).

Details

Registry name

io.github.putervision/world-model-mcp

Version

0.6.1

Auth

No auth

Packages

npm @putervision/world-model-mcp

License

MIT

Repository

github.com/putervision/world-model-mcp

Website

worldmodelmcp.com

First seen

Oct 24, 2026

Last checked

Oct 24, 2026

Source

Official MCP Registrygithub-graphql