MCP Tools
August 7, 2026 · View on GitHub
52 standalone tools register unconditionally; 3 more register only when an upstream MCP server is configured (55 total with both present).
# source: tests_py/test_main.py::test_standalone_baseline_is_52_tools
# verified 2026-07-12 by a live DB-less `tools/list` stdio round-trip
# against `bare-container-contract` + `wiki_migrate`, commit 4be298a3;
# bumped to 51 by `check_setup` (issue #115), to 52 by `ingest_document`
# (offline .docx / Confluence export ingest, issue #192).
Tier 1 — Core Memory & Profiling (22 tools)
| Tool | Purpose | Target Latency |
|---|---|---|
query_methodology | Returns cognitive profile + hot memories for current domain | <50ms |
detect_domain | Lightweight domain classification | <20ms |
rebuild_profiles | Full rescan of session data | <10s |
list_domains | Overview of all domains | <10ms |
record_session_end | Incremental profile update + session critique | <200ms |
explore_features | Interpretability exploration (features, attribution, persona, crosscoder) | <100ms |
remember | Store a memory through the 4-signal predictive coding gate | <100ms |
recall | Retrieve memories via 6-signal WRRF fusion | <200ms |
consolidate | Run maintenance: decay, compression, CLS, sleep compute | <5s |
checkpoint | Save/restore working state for hippocampal replay | <100ms |
narrative | Generate project narrative from stored memories | <500ms |
memory_stats | Memory system diagnostics | <50ms |
import_sessions | Import conversation history into memory store | varies |
forget | Hard/soft delete with is_protected guard; hard delete is cross-substrate (row + derived wiki claims + unreferenced raw artifact) | <50ms |
validate_memory | Validate memories against filesystem state | <500ms |
rate_memory | Useful/not-useful feedback → metamemory confidence | <50ms |
seed_project | 5-stage codebase bootstrap | varies |
anchor | Mark memory as compaction-resistant (heat=1.0) | <50ms |
backfill_memories | Auto-import prior Claude Code conversations | varies |
unified_search | Unified retrieval across memories, wiki, and code graph | <200ms |
get_telemetry | Retrieval and memory-system telemetry metrics | <50ms |
check_setup | Verify local install, backend-aware (SQLite: store open + FS; PostgreSQL: PG driver, DATABASE_URL, connection, extensions, FS) — facade over mcp_server.doctor | <500ms |
Tier 2 — Navigation & Exploration (7 tools)
| Tool | Purpose | Target Latency |
|---|---|---|
recall_hierarchical | Fractal L0/L1/L2 weighted recall | <200ms |
drill_down | Navigate into fractal cluster (L2 → L1 → memories) | <100ms |
navigate_memory | Successor Representation co-access BFS traversal | <200ms |
get_causal_chain | Trace entity relationships through knowledge graph | <200ms |
detect_gaps | Identify isolated entities, sparse domains, temporal drift | <500ms |
recall_skills | Recall learned procedural skills by situation | <200ms |
why | Resolve ⟦rcpt:id⟧ injection receipts into presence-in-context evidence (blame path, decision 4255039) | <100ms |
Tier 3 — Automation & Intelligence (10 tools)
| Tool | Purpose | Target Latency |
|---|---|---|
sync_instructions | Push top memory insights into CLAUDE.md | <500ms |
create_trigger | Prospective memory triggers (keyword/time/file/domain) | <100ms |
add_rule | Add neuro-symbolic hard/soft/tag rules | <100ms |
get_rules | List active rules by scope/type | <50ms |
get_project_story | Period-based autobiographical narrative | <500ms |
assess_coverage | Knowledge coverage score (0-100) + recommendations | <500ms |
codebase_analyze | Native AST codebase analysis (tree-sitter, 7 languages) | varies |
curate_wiki | Auto-curate wiki pages from memory clusters | varies |
curate_distill | Return understanding-level distillation dossiers (error->success, co-access, entity family) for the LLM to author lesson memories from (M-D8) | ~200-500ms |
ingest_document | Ingest a .docx or Confluence storage-format XHTML export into the memory/wiki store, with provenance + idempotent re-ingest (issue #192). File-based, no upstream needed | varies |
Tier 4 — Wiki (10 tools)
| Tool | Purpose | Target Latency |
|---|---|---|
wiki_write | Create a first-class wiki page (ADR, spec, note) | <100ms |
wiki_read | Read a wiki page | <50ms |
wiki_list | List wiki pages by scope/kind | <50ms |
wiki_link | Create a typed link between wiki pages | <50ms |
wiki_adr | Create an Architecture Decision Record | <100ms |
wiki_rename | Rename a wiki page and update backlinks | <100ms |
wiki_verify | Verify wiki page integrity and links | <100ms |
wiki_reindex | Reindex wiki pages into memory pointers | varies |
wiki_purge | Permanently delete a wiki page | <50ms |
wiki_migrate | Reconcile wiki.pages against FS (backfill + ghost purge) | varies |
Upstream-integration tools (3, conditionally registered)
These register only when their upstream MCP server is configured, bringing
the total to 55: ingest_codebase + change_impact (ai-architect-mcp-codebase)
and ingest_prd (ai-architect-mcp-spec). With no upstream present, exactly the
52 standalone tools above register. Driving the ai-architect pipeline
end-to-end (formerly run_pipeline) is not part of this server — it
lives in the ai-architect-mcp-codebase MCP.
Slash Commands
Discovered from commands/*.md at the repo root (not .claude/commands/ —
this is a plugin repo, so the plugin loader picks these up directly). Each
command is a single Markdown file with a name/description frontmatter
pair; /preflight additionally scopes allowed-tools to keep itself
read-only.
| Command | What it does | Roles |
|---|---|---|
/methodology | Retrieves the cognitive methodology profile (via query_methodology) for the current working directory and offers rebuild_profiles / list_domains follow-ups, plus hypermnesia-mcp-viz's get_methodology_graph when that companion MCP is installed | Any user, any session — the general entry point into a domain's profile |
/why | Deterministic blame-path: resolves ⟦rcpt:id⟧ presence-in-context markers via the why tool, reports which memories were in context (never that they caused an answer — Pearl-rung-1 evidence only) | Anyone auditing why an answer looked the way it did |
/preflight [symptôme] | Runs python -m mcp_server.doctor (backend-aware check list) and turns the output into a dependency-ordered, copy-paste repair plan; takes an optional symptom argument to prioritize the relevant check first. Read-only — modifies no files | New users whose install doesn't work yet; support; first-deploy DevOps (issue #119) |
Convention for adding a new command: one new .md file under
commands/ (frontmatter: name, description, and allowed-tools if the
command should run with restricted permissions) plus one new row in the
table above — the catalogue and the command ship in the same commit.
Data Flow
Memory Write Path
- Gate: 4-signal novelty filter (embedding distance, entity overlap, temporal proximity, structural similarity). The content-derived bypasses are granted by an allowlist of origins (
deliberate,local_action— issue #365), so fetched text cannot skip the gate by looking like a decision or an error, and neither can content from a channel nobody has classified;forceand adeliberatewrite class still bypass at any origin. Decision/error content bypasses the gate — detection is language-aware (seedocs/data-flow.md§ Write Gate Bypass);force=trueor animportant/criticaltag always bypasses, in any language - Curate: Active curation — merge with similar, link to related, or create new
- Store: PostgreSQL + pgvector with auto tsvector indexing → entity extraction → knowledge graph
Memory Read Path
- Route: Intent classification (temporal/causal/semantic/entity/knowledge_update/multi_hop)
- Enrich: Doc2Query expansion + concept synonyms
- Fuse: PL/pgSQL
recall_memories()— WRRF fusion of vector + FTS + trigram + heat + recency (server-side) - Rerank: FlashRank cross-encoder (client-side, top-3x candidates)
- Filter: Neuro-symbolic rules → ranked results
Cognitive Profile Pipeline
- Scan: Read ~/.claude/projects/ for JSONL conversations and memory .md files
- Group: Map projects to domains via project ID matching
- Extract: Per-domain pattern extraction (clustering, n-grams, tool stats, session shape)
- Classify: Felder-Silverman cognitive style from behavioral signals
- Bridge: Cross-domain connections from brain-index cross-refs and text analogies
- Detect gaps: Blind spots by comparing domain coverage against global averages
- Learn features: Sparse dictionary learning on 27D behavioral activation space
- Encode: Per-domain sparse feature activations + persona vectors
- Crosscode: Detect persistent behavioral features across domains
- Store: Persist as ~/.claude/methodology/profiles.json