MCP Tools Reference

DepthFusion exposes 38 MCP tools to Claude Code. 25 are always-on; 13 are feature-flagged.

Known drift (E-87/S-337): this count and table predate several epics (S-271 gate-outcome tools, S-293/S-294 model-intelligence tools) and were already stale before this file's event-family rows were added below — the real canonical count is len(TOOLS) in mcp/tools/_registry.py (50 as of S-337; see mcp/server.py's module docstring). Reconciling the rest of this table is tracked separately; do not treat the header count above as accurate.

Core tools (always enabled)

Tool	Description
`depthfusion_status`	Return current DepthFusion component status
`depthfusion_recall_relevant`	Retrieve most relevant session blocks for a query. Args: `query`, `top_k` (default 5), `snippet_len` (default 1500), `cross_project` (default false), `project` (optional), `explain` (bool), `mode` (full/index/timeline)
`depthfusion_tag_session`	Tag a session file with metadata
`depthfusion_auto_learn`	Trigger auto-learning extraction from recent session files
`depthfusion_compress_session`	Compress a specific .tmp session file into a discovery file
`depthfusion_confirm_discovery`	Actively confirm a decision or fact for immediate capture
`depthfusion_set_memory_score`	Override importance and/or salience scalars on a discovery file. Args: `filename`, `importance` (0.0–1.0), `salience` (0.0–5.0)
`depthfusion_recall_feedback`	Apply bounded salience deltas based on which retrieved chunks were actually used. Args: `recall_id`, `used` (chunk_id[]), `ignored` (chunk_id[])
`depthfusion_pin_discovery`	Pin or unpin a discovery file so it is exempt from age-based pruning. Args: `filename`, `pinned` (default true)
`depthfusion_record_telemetry`	Log a per-tool-call telemetry event for cost, latency, and usage analytics. Args: `session_id`, `tool_name`, `duration_ms`, `tokens_in`, `tokens_out`, `cost_usd_estimate`
`depthfusion_query_telemetry`	Aggregate telemetry events by project, agent, story, sprint, or period. Args: `project`, `agent`, `story_id`, `sprint`, `period` (day/week/month)
`depthfusion_session_seed`	Run a seed recall query at session start and publish results as high-priority ContextItems. Args: `session_id`, `top_k`, `mode` (recall/fabric_seed), `projects`, `goal`
`depthfusion_register_project`	Register a local project with DepthFusion for context tracking. Args: `slug`, `name`, `local_path`, `github_url`, `description`
`depthfusion_list_projects`	List all registered projects
`depthfusion_sync_project`	Sync a registered project's BACKLOG.md, CLAUDE.md, and recent git log into the knowledge base. Args: `slug`
`depthfusion_ingest_project`	Ingest a project into the knowledge base from a local path or GitHub URL. Args: `slug`, `source`, `mode` (structural/full)
`depthfusion_research_topic`	Research a topic using web search (DuckDuckGo), arXiv, and GitHub. Args: `topic`, `slug`, `sources`
`depthfusion_bridge`	Delegate a prompt to an external LLM via OpenRouter with shared DepthFusion memory. Args: `model`, `prompt`, `context_tags`
`depthfusion_ingest_conversation`	Bulk-import a past conversation from ChatGPT, Gemini, or DeepSeek. Args: `provider` (chatgpt/gemini/deepseek/generic), `data`
`depthfusion_list_providers`	List configured bridge providers and their status
`depthfusion_recommend_model`	Return a ranked list of models for a given task category. Args: `task_category` (code/review/planning/docs), `exclude_vendors`, `budget_usd`
`depthfusion_describe_capabilities`	Describe which DepthFusion layers and mechanisms are active on this instance
`depthfusion_session_checkpoint`	Publish a manual session checkpoint at a plan boundary, retrievable via `depthfusion_session_seed(mode="resume")`. Args: `session_id`, `project_slug`, `plan_state`, `files_modified`
`depthfusion_index_evidence`	Persist one DMA verification evidence record to the evidence store. Args: `project_slug`, `run_id`, `task_id`, `evidence_type`, `verdict`, `payload`
`depthfusion_recall_evidence`	Query DMA verification evidence records by project and run. Args: `project_slug`, `run_id` (optional)
`depthfusion_event_publish`	Publish content as a MemoryEntity + EventEntity, indexed into HNSW and owned by the authenticated principal's project scope (E-87/S-337; re-registered). Args: `content`, `agent_id`, `project_slug`, `session_id` (optional), `event_type` (optional)
`depthfusion_event_seed`	Return a ranked context bundle across one or more projects for session warm-up (E-87/S-337; re-registered; read-only). Args: `projects`, `goal` (optional), `top_k` (optional), `since_hours` (optional)
`depthfusion_agent_trail`	Return AGENT_PUBLISHED/AGENT_RECEIVED EventEntities for an agent, scoped to the authenticated principal's authorized projects (E-87/S-337; re-registered; read-only). Args: `agent_id`, `project` (optional), `since` (optional), `until` (optional)

Feature-flagged tools

These tools are enabled by specific configuration flags. Check depthfusion_describe_capabilities to see which are active on your instance.

Tool	Flag	Description
`depthfusion_publish_context`	`router_enabled`	Publish a context item to the bus with idempotent dedup by content hash. Args: `item` (item_id, content, source_agent, tags)
`depthfusion_graph_traverse`	`graph_enabled`	Traverse entity graph from a named entity
`depthfusion_graph_status`	`graph_enabled`	Report graph health: node count, edge count, coverage, tier
`depthfusion_set_scope`	`graph_enabled`	Set session graph scope (project / cross_project / global), with optional sub_scope
`depthfusion_get_provenance_graph`	`graph_enabled`	Returns DERIVED_FROM/AGENT_PUBLISHED/PRECEDED_BY subgraph for a memory or session. Args: `memory_id` or `session_id`, `max_depth`
`depthfusion_retrieve_context`	`cognitive_retrieval`	Cognitive retrieval with 8-component scoring and retrieval trace. Args: `query`, `project_id`, `top_k`, `memory_types`
`depthfusion_record_decision`	`decision_memory`	Record an architectural or implementation decision as a typed memory. Args: `project_id`, `decision`, `rationale`, `topic_key`, `rejected_options`, `impact_radius`
`depthfusion_adjudicate_decision`	`decision_memory`	Record that a decision conflict was resolved via explicit adjudication. Args: `memory_id`, `rationale`, `winning_option`
`depthfusion_clarify_decision`	`decision_memory`	Append a clarification to an existing decision record. Args: `memory_id`, `clarification`
`depthfusion_abort_decision`	`decision_memory`	Abort an in-flight decision — archives the memory and emits ABORTED event. Args: `memory_id`, `reason`
`depthfusion_record_incident`	`operational_memory`	Record an error→fix→lesson triple as operational memory. Args: `project_id`, `error`, `fix`, `lesson`, `severity`
`depthfusion_mark_superseded`	`operational_memory`	Mark a memory as superseded by a newer one. Args: `project_id`, `old_memory_id`, `new_memory_id`, `reason`
`depthfusion_report_outcome`	`operational_memory`	Record the outcome of applying a decision or procedure. Args: `project_id`, `memory_id`, `outcome`, `success`

Enabling feature-flagged tools

Set the corresponding env var or profile to activate a flag group:

# Graph tools
DEPTHFUSION_GRAPH_ENABLED=true

# Cognitive retrieval
DEPTHFUSION_COGNITIVE_SCORING=true

# Decision + operational memory
DEPTHFUSION_DECISION_MEMORY=true

# Or use the research profile (enables all):
DEPTHFUSION_PROFILE=research

See Configuration for the full profile reference.