MCP Tools Model Context Protocol tools reference
the-brain exposes these tools via MCP for use with Claude Desktop, Cursor, Zed, and other MCP clients.
Tool Description memory_searchSearch memories by query memory_storeStore new memory memory_contextGet context for a prompt memory_listList paginated memories
Tool Description graph_searchSearch graph nodes graph_add_nodeCreate graph node graph_connectConnect two nodes
Tool Description brain_statsGet comprehensive statistics brain_config getRead a config value brain_config setSet a config value identity_getGet identity anchor data identity_updateUpdate identity traits
Tool Description project_listList known projects project_switchSwitch active project
Tool Description training_statusCheck training state training_consolidateRun consolidation
Tool Description pipeline_ingestIngest raw content pipeline_statusPipeline queue status
Tool Description scheduler_listList scheduled tasks scheduler_scheduleCreate task scheduler_cancelRemove task
Tools for integrating the-brain as a cognitive layer in meta-harness systems (AHE, Meta-Harness). Enables predictive regression detection across harness evolution cycles.
Tool Description brain_predict_regressionPredict expected benchmark score ranges before harness edits. Returns per-metric confidence intervals. brain_record_runRecord benchmark results after evaluation. Updates fingerprints and returns surprise assessment per metric. brain_get_fingerprintGet per-model per-benchmark performance fingerprints — running mean, std, sample count. brain_get_regression_graphGet causal graph of harness edits → benchmark regressions from graph memory. brain_get_surprise_feedGet anomalous results for HITL review — results where observed score deviated >2σ from baseline. brain_compare_agentsCompare multiple models on a benchmark — rankings with per-metric differences.
AHE Proposer generates edit → brain_predict_regression("claude", "mmlu") ← "acc: 0.8900–0.9100 (85% confidence)" AHE Evaluator runs benchmarks → brain_record_run("claude", "mmlu", {acc: 0.85}, "edit-247") ← "⚠️ acc: 0.8500 vs expected 0.8900–0.9100, z=3.20" HITL Reviewer checks anomalies → brain_get_surprise_feed(min_surprise: 0.7) ← "claude/mmlu/acc: z=3.20, claude/gsm8k: z=2.85" Next cycle proposer looks for patterns → brain_get_regression_graph(model: "claude") ← "Edit #89 (tool registry) also regressed Claude on MMLU" URI Description brain://statsBrain statistics brain://memories/recentRecent memories brain://graph/high-weightHigh-weight nodes brain://identityIdentity anchor brain://projectsProject list brain://training/statusTraining state brain://configCurrent config