agentmako
agentmako is a local-first codebase intelligence engine that gives AI coding agents a rich, evidence-backed tool surface for understanding, navigating, and verifying a project before editing it. Key capabilities include:
Core Query & Search
reef_ask– Primary query engine: ask any project question and receive compiled code + database evidence packetscontext_packet– Turn a request into ranked, source-labeled contextlive_text_search/ast_find_pattern/cross_search– Live filesystem, structural AST, and cross-source search
Route, File & Symbol Intelligence
route_trace,file_health,auth_path,schema_usage,trace_file,trace_table,trace_rpc,trace_edge,trace_error– End-to-end tracing of routes, files, tables, RPCs, and errorssymbols_of/exports_of– List symbols declared or exported by a filepreflight_table– Full surface for a table (columns, indexes, FK, RLS, triggers, routes)
Import Graph & Dependency Analysis
imports_deps,imports_impact,imports_hotspots,imports_cycles– Map dependencies, impact, hotspots, and circular references
Graph & Flow Tools
graph_neighbors,graph_path,flow_map,change_plan– Traverse entity graphs, find paths, map flows, and plan changes
Neighborhood Composers
table_neighborhood,route_context,rpc_neighborhood– Combined context for tables, routes, and RPCs in one call
Live Database Inspection
db_ping,db_columns,db_fk,db_rls,db_rpc,db_table_schema,db_reef_refresh,db_review_comment– Inspect and annotate live DB schema, RLS, and procedures
Diagnostics
typescript_diagnostics,eslint_diagnostics,oxlint_diagnostics,biome_diagnostics,lint_files,git_precommit_check,diagnostic_refresh– Run linters and type-checkers with persisted findings
Reef Facts, Findings & Verification
project_findings,file_findings,project_facts,file_preflight,reef_verify,reef_status,reef_known_issues,project_open_loops,verification_state– Query durable findings/facts, verify completion, and surface open issuesreef_diff_impact,reef_overlay_diff,evidence_confidence,evidence_conflicts– Assess impact of changes and surface stale or contradictory evidence
Rules, Conventions & Learning
project_conventions,list_reef_rules,rule_pack_validate,extract_rule_template,reef_learning_review,reef_instructions– Surface and manage project conventions, rules, and learned patterns
Workflow & Planning
mako_help,suggest,investigate,workflow_packet,ask– Orchestrate workflows, get tool recommendations, and run bounded read-only investigations
Artifact Generators
task_preflight_artifact,implementation_handoff_artifact,review_bundle_artifact,verification_bundle_artifact– Compose structured implementation briefs, handoffs, review bundles, and verification plans
Security Audits
tenant_leak_audit– Audit tenant-keyed tables and RLS postureowasp_audit– OWASP Top 10 (2025) scan with CWE references
Session & Feedback
session_handoff,recall_answers,recall_tool_runs,agent_feedback,health_trend,issues_next– Track session state, recall prior runs, rate tool outputs, and monitor project health trends
Index & Project Management
project_index_status,project_index_refresh,repo_map,tool_batch,tool_search,file_write– Manage the project index, emit repo outlines, batch lookups, and write files (with approval)
Ingests Biome diagnostics.
Ingests ESLint diagnostics and review notes.
Ingests staged git diagnostics and provides pre-commit checks.
Provides optional Postgres/Supabase schema snapshots and read-only database inspection with review comments.
Ingests TypeScript diagnostics.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@agentmako give me a context packet for the login flow"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
agentmako
agentmako is a local-first codebase intelligence engine for AI coding tools.
It gives agents like Codex, Claude Code, Cursor, and local harnesses a compact Reef-first tool surface for understanding a project before they edit it. Mako indexes your repo, builds local SQLite-backed facts, tracks diagnostics and review notes, and answers evidence-backed questions instead of making the agent orchestrate broad tool chains or rediscover everything with raw grep.
Mako is built for the first mile of coding-agent work:
What files matter? What routes, symbols, tables, diagnostics, and prior findings are relevant? What should the agent read next?
What You Get
MCP server for coding agents:
agentmako mcpLocal dashboard:
agentmako dashboardPrimary project query:
reef_askacross code, database, findings, diagnostics, instructions, freshness, and literal checksQueryable workflow orientation:
mako_helpDeterministic context expansion:
context_packet_hintson tool results so agents get result-specific next stepsCentral MCP annotations so clients can distinguish safe reads, live reads, and local-state mutations
Compact loop/fallback tools:
reef_status,reef_verify,reef_impact,live_text_search,lint_files, andtool_batchReef Engine facts and findings across indexed, working-tree, and staged state
Reef convention extraction for auth guards, runtime boundaries, generated paths, route patterns, and schema usage
TypeScript, ESLint, Oxlint, Biome, and staged git diagnostic ingestion
Hot-reloaded
.mako/rulesYAML rule packs, including primitive cross-file helper-bypass rules viacanonicalHelperOptional Postgres/Supabase schema snapshots and read-only DB inspection
Local DB review comments for notes on tables, RLS, triggers, publications, subscriptions, and replication
Recall, acknowledgements, and agent feedback for repeated review work
Everything important runs locally. No hosted service is required.
Related MCP server: Atlas
Install
Requires Node.js 20 or newer.
npm install -g agentmakoConfirm the CLI is available:
agentmako --version
agentmako doctorYou should see green checks for configuration and the local API service.
Prefer to build from source (e.g. to contribute)? See Develop From Source at the bottom of this file.
Happy Path Setup
1. Attach your real project
Go to the project you want Mako to understand:
cd C:/path/to/your/projectAttach and index it:
agentmako connect . --no-dbUse --no-db for the first run. It gets the code intelligence path
working before adding database scope.
2. Confirm Mako sees the project
agentmako status .
agentmako tool listRun a real Reef query:
agentmako --json tool call . reef_ask "{\"question\":\"where should I inspect auth route state?\"}"If that returns an evidence-backed answer, facts, findings, or next queries, the core setup is working.
reef_ask plans over code, database, durable findings, diagnostics, and exact
literal evidence. App-flow questions favor file, route, and finding evidence;
RLS/schema questions favor database facts and review comments. To inspect
project rules of thumb directly:
agentmako --json tool call . project_conventions "{}"3. Configure your MCP client
Add this to your MCP client config:
{
"mcpServers": {
"mako-ai": {
"command": "agentmako",
"args": ["mcp"]
}
}
}Restart the MCP client and confirm the mako-ai server starts.
In the agent, default to reef_ask. The compact starting surface is:
reef_askfor project questions across code, database, findings, diagnostics, freshness, and quoted literal checksreef_statusfor maintained issues, changed files, stale diagnostics, and watcher/schema healthreef_verifyfor the completion gate over diagnostic freshness and open loopsreef_impactfor changed-file blast radius and convention risksmako_helpfor an ordered workflow recipe with prefilled argumentslive_text_searchfor exact current-disk regex/glob inventorieslint_filesfor bounded diagnostics and.mako/rulesfindingstool_batchfor independent read-only follow-upstool_searchto discover specialized route, graph, DB, finding, refresh, or context-expansion tools only when the compact surface points at a concrete need
4. Optional: use an agent plugin
Plain MCP works anywhere, but the bundled plugins add Mako-specific skills and
include the same agentmako mcp wiring.
Prerequisites:
Claude Code installed
Node.js 20+ on
PATH(the plugin runsnpx -y agentmako mcp, which fetches the publishedagentmakopackage automatically — no separate global install required)Your target project already attached with
agentmako connect
Claude Code stable path:
claude plugin validate .\mako-ai-claude-plugin
claude --plugin-dir .\mako-ai-claude-pluginNew generated plugin layouts:
claude plugin validate ./plugins/claude-code
codex marketplace add ./plugins
ln -s "$(pwd)/plugins/cursor" ~/.cursor/plugins/local/mako-ai
gemini extensions install ./plugins/geminiInside the agent, confirm the mako-ai MCP server is connected.
The plugin exposes these skills:
/mako-ai:mako-guide/mako-ai:mako-discovery/mako-ai:mako-trace/mako-ai:mako-neighborhoods/mako-ai:mako-graph/mako-ai:mako-database/mako-ai:mako-code-intel/mako-ai:mako-workflow
Use the plugin when you want Claude Code to load Mako-specific guidance for which tools to call and how to interpret their results.
5. Optional: launch the dashboard
From your target project:
agentmako dashboard .This starts the local API, harness service, and web dashboard.
6. Optional: add Supabase/Postgres awareness
Mako works without a database. Add this only after code intelligence is working.
For a one-time interactive setup:
agentmako connect .For CI or scripted setup using an environment variable:
set DATABASE_URL=postgres://...
agentmako connect . --db-env DATABASE_URL --yesThen refresh and verify the local schema snapshot:
agentmako refresh .
agentmako verify .Interactive mode stores database secrets in your OS keychain by default. Project config stores references, not plaintext DB URLs.
Normal Daily Loop
From the target project:
agentmako status .
agentmako dashboard .
agentmako --json tool call . context_packet "{\"query\":\"fix the broken auth callback route\"}"For staged review checks:
agentmako git precommit . --jsonFor database review notes:
agentmako --json tool call . db_review_comment "{\"objectType\":\"replication\",\"objectName\":\"supabase_database_replication\",\"category\":\"review\",\"comment\":\"Check publication coverage before relying on realtime events.\",\"tags\":[\"supabase\",\"replication\"]}"Develop From Source
If you want to hack on Mako itself, clone and build instead of installing from npm.
Prerequisites:
Node.js 20 or newer
Git
Corepack (
corepack enable, included with modern Node.js)
git clone https://github.com/drhalto/agentmako.git
cd agentmako
corepack pnpm install
corepack pnpm run build
npm link ./apps/clinpm link ./apps/cli makes the source-built CLI available as
agentmako on your PATH, replacing any global npm install. Re-run
corepack pnpm run build after pulling changes.
To go back to the published version: npm install -g agentmako.
Development Checks
corepack pnpm run typecheck
corepack pnpm run build
corepack pnpm run test:smoke:reef-tooling
corepack pnpm run test:smoke:reef-model-facing-viewsFull verification:
corepack pnpm testRepository Layout
apps/
cli/ agentmako CLI and MCP entrypoint (the published package)
web/ local dashboard
packages/
contracts/ public TypeScript contracts and tool schemas
config/ shared config helpers
logger/ shared logger
sdk/ programmatic SDK
store/ SQLite stores, migrations, and query helpers
tools/ shared tool implementations
harness-core/ local agent harness runtime
harness-tools/ action tools available to the harness
harness-contracts/ harness contracts and provider catalog
services/
api/ local API and MCP transports
engine/ Reef Engine fact/finding pipeline
harness/ local harness HTTP service
indexer/ repo and schema indexing logic
worker/ background worker
extensions/ provider and integration packages
storage/ schema migrations, models, queries
test/smoke/ smoke coverage
mako-ai-claude-plugin/ Claude Code plugin with Mako skillsMore Docs
License
Apache-2.0. See LICENSE.
Available Tools
105 toolsagent_feedbackagent_feedbackA
Append-only feedback tool for rating a prior Mako tool run from the agent perspective. Use sparingly when a result was notably good, notably bad, or wrong; do not emit routine feedback after every call. Requires referencedToolName and referencedRequestId so each row is tied to a specific prior run. Grade semantics: full = helped complete the task, partial = somewhat useful but flawed, no = wrong or wasted the turn. Starter reason codes: full: answer_complete, evidence_sufficient, trust_matches; partial: partial_coverage, noisy, stale_evidence, missing_known_caller, top_not_useful; no: answer_wrong, wasted_turn, tool_did_nothing, schema_missing.
Inputs: grade (required), reasonCodes (required), referencedToolName (required), referencedRequestId (required), +3 more. Returns: _hints, eventId, toolName, projectId, +1 more
agent_feedback_reportagent_feedback_reportARead-onlyIdempotent
Read-only inspection over agent_feedback rows in mako_usefulness_events: return grade counts by referenced tool plus bounded recent entries, filterable by referencedToolName, grade, and ISO time window. Use to review which tools agents found helpful, noisy, wrong, or incomplete.
Inputs: grade, limit, since, until, +3 more. Returns: _hints, byTool, entries, toolName, +4 more
apply_patchADestructive
Apply a multi-file unified diff. Hunks must apply cleanly with exact context.
Inputs: diff (required). Returns: ok, error, _hints, requiresHarnessSession
askaskARead-only
Router tool: map one natural-language engineering question to one canonical named tool, or conservatively fall back to free_form.
Inputs: question (required), projectId, projectRef. Returns: mode, _hints, result, toolName, +5 more
ast_find_patternast_find_patternARead-onlyIdempotent
Code-intelligence tool for structural pattern search: run one ast-grep pattern against every fresh indexed TypeScript / TSX / JavaScript / JSX file and return typed matches (file, line range, column range, matched text, captured metavariables). Ambiguous TSX/JSX snippets that start with {, [, or < retry with an auto-anchored const _ = ... parser context when the original pattern returns zero matches, and patternAttempts / match patternVariant report which form matched. Reef-backed freshness guard skips stale/deleted/unknown indexed files so old AST rows do not become phantom matches; set MAKO_REEF_BACKED=legacy for the one-release rollback path. Read-only; never edits files. Use when text/FTS search is too noisy for structural queries (e.g. console.log($X), await supabase.rpc($NAME), useEffect($FN, [])).
Inputs: pattern (required), captures, maxFiles, pathGlob, +5 more. Returns: _hints, matches, pattern, toolName, +9 more
auth_pathauth_pathARead-onlyIdempotent
Answer tool for auth questions: trace likely auth boundaries for a route, file, or feature without overclaiming. When no exact route/file/feature match exists, returns matched:false with suggestedNext:{ tool:"cross_search", args:{...} } instead of throwing, so batches and agent workflows can continue.
Inputs: file, route, feature, projectId, +1 more. Returns: _hints, reason, result, matched, +4 more
biome_diagnosticsbiome_diagnosticsA
Explicit Reef ingestion tool for Biome diagnostics: run the project's local Biome executable or GitLab-reporter package script on requested project-relative files with check --reporter=gitlab, persist working-tree ProjectFinding rows under source biome, and record unavailable/error/succeeded run status. File-mode only; deliberately avoids Biome's experimental JSON reporter.
Inputs: files (required), projectId, projectRef, scriptName, +1 more. Returns: _hints, status, command, exitCode, +12 more
change_planchange_planCRead-onlyIdempotent
Graph tool for bounded change scope: return path-derived direct surfaces, one-hop dependent surfaces, and an explicit dependency order.
Inputs: startEntity (required), targetEntity (required), direction, edgeKinds, +4 more. Returns: _hints, result, toolName, projectId
context_packetcontext_packetARead-onlyIdempotent
Context scout for coding agents: turn a messy request into ranked, source-labeled, readable context using deterministic providers first (bounded live quoted-literal search, files, routes, symbols, schema, import graph, centrality, hot hints). Retrieval diagnostics expose providerExecutionMode, totalProviderDurationMs, slowestProvider, evidence gates, and executable follow-ups so agents can distinguish recall gaps from slow or incomplete provider lanes. Reef-backed enrichments add working-tree overlay metadata and active findings; use risksMinConfidence to suppress low-confidence risk speculation. Set MAKO_REEF_BACKED=legacy for the one-release rollback path. Read-only; does not refresh the index. Use as the first-mile packet before normal harness read/search/edit loops.
Inputs: request (required), mode, projectId, focusFiles, +13 more. Returns: mode, risks, _hints, intent, +23 more
create_fileADestructive
Create a new file at a project-relative path. Errors if the file already exists; use file_write to overwrite.
Inputs: path (required), content (required). Returns: ok, error, _hints, requiresHarnessSession
cross_searchcross_searchARead-onlyIdempotent
Search a term across code chunks, schema objects, RPC/trigger bodies, routes, and stored memories in one call. Exact code literals route to a bounded live_text_search preview. Use live_text_search directly for full inventories, regex, or custom glob scope.
Inputs: term (required), limit, projectId, verbosity, +1 more. Returns: _hints, result, toolName, projectId
db_columnsdb_columnsBRead-onlyIdempotent
Read-only database tool for column questions: inspect only columns and primary-key details for a table via pg_catalog.
Inputs: table (required), schema, projectId, projectRef. Returns: table, _hints, schema, columns, +1 more
db_fkdb_fkARead-onlyIdempotent
Read-only database tool for relationship questions: inspect inbound and outbound foreign-key references for a table via pg_catalog.
Inputs: table (required), schema, projectId, projectRef. Returns: table, _hints, schema, inbound, +2 more
db_pingdb_pingARead-onlyIdempotent
Read-only database tool: verify connectivity and surface platform, version, schemas, and transaction state.
Inputs: projectId, projectRef. Returns: _hints, schemas, database, platform, +5 more
db_reef_refreshdb_reef_refreshA
Explicit Reef DB ingestion tool: replace indexed Reef facts for the current schema snapshot, including schemas, tables, columns, indexes, foreign keys, RLS policies, triggers, enums, RPCs, scheduled jobs, RPC-to-table refs, and indexed schema usages. Uses the existing schema snapshot/read-model indexes; run project_index_refresh first if the snapshot is stale or missing.
Inputs: freshen, projectId, factsLimit, projectRef, +2 more. Returns: facts, _hints, summary, toolName, +5 more
db_review_commentdb_review_commentA
Append-only DB review note tool: leave a short AI/operator comment on a database object such as a table, column, policy, trigger, publication, replication slot, or general replication topic. Writes only to Mako's local project store; it never mutates the live database.
Inputs: comment (required), objectName (required), objectType (required), tags, +9 more. Returns: _hints, comment, preview, toolName, +4 more
db_review_commentsdb_review_commentsARead-onlyIdempotent
Read DB review comments from Mako's local append-only ledger, filterable by database object, category, tag, free-text query, or target fingerprint. Use during schema/table/RLS/replication review to recover prior AI/operator notes.
Inputs: tag, limit, query, category, +7 more. Returns: _hints, filters, comments, toolName, +4 more
db_rlsdb_rlsARead-onlyIdempotent
Read-only database tool for security questions: inspect row-level security state and policies for a table via pg_catalog.
Inputs: table (required), schema, projectId, projectRef. Returns: table, _hints, schema, forceRls, +3 more
db_rpcdb_rpcARead-onlyIdempotent
Read-only database tool for function questions: inspect one stored procedure/function signature by name, or pass list: true to enumerate routines with signatures, return shape, language, and security.
Inputs: list, name, limit, schema, +5 more. Returns: args, mode, name, rpcs, +11 more
db_table_schemadb_table_schemaARead-onlyIdempotent
Read-only database tool for full table shape: inspect columns, indexes, constraints, foreign keys, RLS, and triggers.
Inputs: table (required), schema, projectId, projectRef. Returns: rls, table, _hints, schema, +6 more
delete_fileADestructive
Delete a file at a project-relative path. Snapshot captures the bytes for undo.
Inputs: path (required). Returns: ok, error, _hints, requiresHarnessSession
diagnostic_refreshdiagnostic_refreshA
Explicit Reef diagnostic ingestion runner: invoke selected diagnostic sources (lint_files, programmatic_findings, typescript_syntax, TypeScript, ESLint, Oxlint, Biome, or git_precommit_check), persist their Reef diagnostic runs/findings through the underlying tools, and return a compact per-source summary. File-mode sources are skipped unless files are supplied; this is a mutation tool because it refreshes Reef state.
Inputs: files, scripts, sources, projectId, +5 more. Returns: _hints, results, summary, findings, +4 more
eslint_diagnosticseslint_diagnosticsA
Explicit Reef ingestion tool for ESLint diagnostics: run the project's local ESLint executable on the requested project-relative files with JSON output, persist working-tree ProjectFinding rows under source eslint, and record unavailable/error/succeeded run status. File-mode only; does not run broad project lint.
Inputs: files (required), projectId, projectRef, scriptName, +1 more. Returns: _hints, status, command, exitCode, +12 more
evidence_confidenceevidence_confidenceCRead-onlyIdempotent
Reef 10 evidence-confidence view: label facts and findings as verified_live, fresh_indexed, stale_indexed, fuzzy_semantic, historical, contradicted, or unknown so model-facing tools can prefer trustworthy evidence.
Inputs: limit, filePath, projectId, projectRef, +1 more. Returns: items, _hints, summary, toolName, +4 more
evidence_conflictsevidence_conflictsARead-onlyIdempotent
Reef 10 conflict view: surface stale indexed evidence, explicit conflict facts, and findings that report incorrect or contradictory evidence, with suggested cross-check actions.
Inputs: limit, filePath, projectId, projectRef, +2 more. Returns: _hints, toolName, warnings, conflicts, +4 more
exports_ofexports_ofBRead-onlyIdempotent
Symbols tool for exports: list only the indexed symbols that a file exports.
Inputs: file (required), projectId, projectRef. Returns: file, _hints, exports, toolName, +4 more
extract_rule_templateextract_rule_templateARead-onlyIdempotent
Read-only Reef rule-pack mining tool: inspect a local git fix diff and propose .mako/rules YAML templates from removed TS/JS anti-pattern shapes, with related durable findings when available. Does not write files or mutate Reef state.
Inputs: fixCommit (required), filePath, projectId, baseCommit, +4 more. Returns: _hints, summary, toolName, warnings, +8 more
file_editADestructive
Replace a substring in an existing file. The substring must occur exactly once unless replaceAll is true.
Inputs: path (required), newString (required), oldString (required), replaceAll. Returns: ok, error, _hints, requiresHarnessSession
file_factsfile_factsARead-onlyIdempotent
Reef read tool for one file's durable facts: computes the file subject fingerprint and returns facts such as working_tree_overlay file_snapshot rows. Use this to inspect what Reef currently knows about a file across overlays.
Inputs: filePath (required), kind, limit, source, +4 more. Returns: facts, _hints, filters, filePath, +6 more
file_findingsfile_findingsARead-onlyIdempotent
Reef read tool for one project-relative file: return durable Reef findings attached to that file, with the same overlay/source/status filters as project_findings, including source aliases for bare rule IDs and rule_pack:. Use this before editing a file when a shell or agent needs known active diagnostics without rerunning broad lint.
Inputs: filePath (required), limit, source, status, +5 more. Returns: _hints, filters, filePath, findings, +6 more
file_healthfile_healthCRead-onlyIdempotent
Answer tool for file questions: summarize a file's role, dependents, and notable risks with evidence.
Inputs: file (required), projectId, projectRef. Returns: _hints, result, toolName, projectId
file_preflightfile_preflightARead-onlyIdempotent
Reef pre-edit composer for one file: returns durable findings, file-scoped diagnostic freshness, source-filtered recent runs, watcher diagnostic state, applicable conventions, and finding acknowledgement history in one read-only packet. Use before editing a file when an agent needs the operational gate without separate file_findings, verification_state, project_conventions, and finding_acks_report calls.
Inputs: filePath (required), sources, ackLimit, projectId, +6 more. Returns: _hints, filters, summary, filePath, +9 more
file_writeADestructive
Create or overwrite a file at a project-relative path. Returns a snapshot id for undo.
Inputs: path (required), content (required). Returns: ok, error, _hints, requiresHarnessSession
finding_ackfinding_ackA
Mutation tool for verified-safe markers: append one row to the finding_acks ledger so subsequent ast_find_pattern / lint_files callers that opt into the same category filter the match out. Category is caller-owned (for lint findings, finding.code is the recommended default). For ast_find_pattern matches, pass match.ackableFingerprint; for lint_files findings, pass finding.identity.matchBasedId. Append-only; duplicate (projectId, category, fingerprint) inserts persist as separate rows and dedupe at query time. Emits one RuntimeUsefulnessEvent per successful call.
Inputs: reason (required), category (required), fingerprint (required), subjectKind (required), +10 more. Returns: ack, _hints, preview, toolName, +2 more
finding_ack_batchfinding_ack_batchA
Mutation batch for reviewed finding acknowledgements: append many finding_acks rows in one call while sharing batch-level defaults for category, subjectKind, status, reason, acknowledgedBy, sourceToolName, and sourceRuleId. Use when a scan returns many reviewed false positives or accepted tradeoffs. Emits one finding_ack telemetry event per successful row.
Inputs: rows (required), reason, status, preview, +8 more. Returns: acks, _hints, preview, summary, +5 more
finding_acks_reportfinding_acks_reportARead-onlyIdempotent
Read-only inspection over the finding_acks ledger: return aggregate counts (by category, status, subjectKind, filePath) and a bounded reverse-chronological list of acks, filterable by category, subjectKind, filePath, status, and ISO time window. Use to see what operators have suppressed and which categories carry the most weight. Never writes to the ledger.
Inputs: limit, since, until, status, +5 more. Returns: acks, _hints, byStatus, toolName, +7 more
flow_mapflow_mapARead-onlyIdempotent
Graph tool for end-to-end flow questions: turn one graph path into ordered flow steps, transitions, and major boundary crossings.
Inputs: startEntity (required), targetEntity (required), direction, edgeKinds, +4 more. Returns: _hints, result, toolName, projectId
git_precommit_checkgit_precommit_checkA
Git pre-commit guard for staged TS/TSX files: reads staged blobs, checks API route auth guards and Next.js server/client boundary mistakes, persists staged Reef findings unless MAKO_REEF_BACKED disables the migration, and returns hook-friendly continue/stopReason output. Uses discovered project-profile auth guards/server-only modules plus optional .mako/git-guard.json/input allowlists. Never edits the index or worktree.
Inputs: projectId, projectRef, authGuardSymbols, publicRouteGlobs, +2 more. Returns: _hints, policy, gitRoot, continue, +10 more
graph_neighborsgraph_neighborsARead-onlyIdempotent
Graph tool for adjacency questions: traverse outward from one or more start entities with explicit direction, depth, and edge filters.
Inputs: startEntities (required), limit, direction, edgeKinds, +5 more. Returns: _hints, result, toolName, projectId
graph_pathgraph_pathARead-onlyIdempotent
Graph tool for connection questions: find one shortest typed path between two entities while keeping heuristic edges explicit and opt-in.
Inputs: startEntity (required), targetEntity (required), direction, edgeKinds, +5 more. Returns: _hints, result, toolName, projectId
health_trendhealth_trendARead-onlyIdempotent
Operator tool for project-state trend review: compare a recent trace window against a prior window without fabricating trend lines when history is thin.
Inputs: limit, projectId, projectRef. Returns: _hints, result, toolName, projectId
implementation_handoff_artifactimplementation_handoff_artifactBRead-only
Artifact tool for reusable implementation handoff: compose one implementation brief and the current session handoff into a typed handoff artifact. Basis is session-scoped — session_handoff shifts with every tool call, so artifactId moves between calls by design (each call captures the current session state). Do not dedupe on artifactId.
Inputs: queryKind (required), queryText (required), export, projectId, +4 more. Returns: _hints, result, exported, toolName, +1 more
imports_cyclesimports_cyclesBRead-onlyIdempotent
Imports tool for cycle detection: detect circular dependencies in the indexed internal import graph.
Inputs: projectId, projectRef. Returns: _hints, cycles, toolName, projectId, +1 more
imports_depsimports_depsARead-onlyIdempotent
Imports tool for direct dependencies: list a file's indexed imports and flag unresolved internal edges.
Inputs: file (required), projectId, projectRef. Returns: file, _hints, imports, toolName, +5 more
imports_hotspotsimports_hotspotsBRead-onlyIdempotent
Imports tool for graph hotspots: rank the most connected files in the indexed internal import graph.
Inputs: limit, projectId, projectRef. Returns: limit, _hints, hotspots, toolName, +2 more
imports_impactimports_impactCRead-onlyIdempotent
Imports tool for downstream impact: trace which indexed files depend on a file through the import graph.
Inputs: file (required), depth, projectId, projectRef. Returns: file, depth, _hints, toolName, +5 more
investigateinvestigateARead-onlyIdempotent
Workflow tool for bounded read-only investigation: run a short sequential tool chain over shipped workflows and return typed step history.
Inputs: question (required), budget, direction, projectId, +5 more. Returns: _hints, result, toolName, projectId
issues_nextissues_nextCRead-onlyIdempotent
Operator tool for queue-oriented recommendations: derive one current issue plus queued follow-on issues from recent unresolved project traces.
Inputs: limit, projectId, projectRef. Returns: _hints, result, toolName, projectId
lint_fileslint_filesC
Code-intelligence tool for diagnostics on an indexed file set: run rule-packs, including canonicalHelper producer/consumer checks, plus TS-aware and structural alignment diagnostics; persist indexed Reef findings/run metadata under source lint_files; and return typed AnswerSurfaceIssue findings. Same engine that powers the answer loop and review_bundle, so findings stay consistent across surfaces.
Inputs: files (required), projectId, verbosity, projectRef, +3 more. Returns: _hints, findings, toolName, warnings, +5 more
list_reef_ruleslist_reef_rulesARead-onlyIdempotent
Reef read tool for rule metadata: list durable rule descriptors, optionally filtered by source namespace or enabled-by-default status. Descriptors explain which facts a finding source consumes and what rule ids mean.
Inputs: limit, projectId, projectRef, enabledOnly, +1 more. Returns: rules, _hints, filters, toolName, +4 more
live_text_searchlive_text_searchARead-onlyIdempotent
Raw live filesystem text search: run ripgrep over the attached project root and return bounded file/line matches. Defaults to fixed-string matching; pass fixedStrings: false to use ripgrep regex syntax. Read-only and explicitly live. Prefer reef_ask for bounded quoted literal questions; use this when you need regex, custom glob scope, or raw full match inventories.
Inputs: query (required), maxFiles, pathGlob, projectId, +5 more. Returns: query, _hints, matches, toolName, +5 more
mako_helpmako_helpARead-onlyIdempotent
Orientation tool for coding agents: given a natural-language task plus optional focusFiles, changedFiles, focusRoutes, focusSymbols, focusDatabaseObjects, route, table, or rpc anchors, return the recommended Mako workflow recipe as an ordered tool sequence with pre-filled suggestedArgs, batchable follow-ups, and notes. Use before reading long AGENTS/CLAUDE docs when deciding how to start with reef_ask and which specialist follow-ups to load. Read-only and does not execute the returned steps.
Inputs: task (required), rpc, route, table, +8 more. Returns: task, notes, steps, _hints, +5 more
owasp_auditowasp_auditARead-onlyIdempotent
Operator tool for OWASP Top 10 (2025) review: scan indexed TS/JS files for evidence-backed security issues mapped to OWASP categories (A01 broken access control, A02 misconfiguration, A04 crypto failures, A05 injection, A07 auth failures, A10 exceptional conditions) with CWE references and a direct_evidence/weak_signal honesty strength. Always returns a full 10-category coverage section, explicitly naming categories it does not statically check (A03 supply chain, A06 insecure design, A08 integrity, A09 logging). Advisory and heuristic — requires acknowledgeAdvisory:true; not a replacement for dedicated SAST/SCA. Read-only; does not persist findings.
Inputs: acknowledgeAdvisory (required), freshen, maxFiles, projectId, +4 more. Returns: _hints, result, toolName, projectId
oxlint_diagnosticsoxlint_diagnosticsA
Explicit Reef ingestion tool for Oxlint diagnostics: run the project's local Oxlint executable or JSON package script on requested project-relative files with --format json, persist working-tree ProjectFinding rows under source oxlint, and record unavailable/error/succeeded run status. File-mode only; does not run broad project lint.
Inputs: files (required), projectId, projectRef, scriptName, +1 more. Returns: _hints, status, command, exitCode, +12 more
preflight_tablepreflight_tableBRead-onlyIdempotent
Return the full preflight surface for a table: columns, primary key, indexes, foreign keys, RLS state + policies, triggers, related routes, and zod schemas whose surrounding file references the table. Snapshot-strict.
Inputs: table (required), schema, projectId, projectRef. Returns: _hints, result, toolName, projectId
project_conventionsproject_conventionsBRead-onlyIdempotent
Reef 9 convention view: surface explicit convention facts plus profile/index/rule-derived convention candidates such as auth guards, runtime boundaries, generated paths, route patterns, and schema usage conventions.
Inputs: kind, limit, status, projectId, +1 more. Returns: _hints, toolName, warnings, projectId, +3 more
project_diagnostic_runsproject_diagnostic_runsARead-onlyIdempotent
Reef read tool for diagnostic source runs: list recent lint/type adapter runs by source and status, including unavailable/error/succeeded state, duration, counts, command, config, cwd, metadata, and derived cache freshness/age. Use this to distinguish no findings from a diagnostic source that did not run or ran too long ago.
Inputs: limit, source, status, projectId, +2 more. Returns: runs, _hints, filters, toolName, +4 more
project_factsproject_factsARead-onlyIdempotent
Reef read tool for durable project facts: query facts by overlay, source, kind, and subject fingerprint. Use this when an agent or shell needs the calculated substrate behind findings without rerunning analysis.
Inputs: kind, limit, source, overlay, +4 more. Returns: facts, _hints, filters, toolName, +5 more
project_findingsproject_findingsARead-onlyIdempotent
Reef read tool for active project findings: query the durable Reef findings view by overlay, source, status, and resolved inclusion. The source filter accepts a producer source, bare rule ID, or rule_pack:. Acknowledged status is derived from the existing finding_ack ledger, so this is the canonical read path for Reef-native lint/rule findings.
Inputs: limit, source, status, overlay, +4 more. Returns: _hints, filters, findings, toolName, +5 more
project_index_refreshproject_index_refreshA
Mutation tool for index freshness: run the project indexer through the current project-store cache when the snapshot is stale, unknown, deleted, or has unindexed files, then return before/after freshness summaries and the new index run. In if_stale mode, unknown freshness refreshes defensively and unindexed files count as stale work.
Inputs: mode, reason, projectId, projectRef. Returns: run, after, stats, _hints, +8 more
project_index_statusproject_index_statusARead-onlyIdempotent
Project tool for index freshness: compare indexed file rows against live disk metadata, report Reef working-tree fact freshness, watch state, latest index run, unindexedScan status, and a suggested next action. By default this checks indexed rows only and reports that the new-file scan was skipped; pass includeUnindexed: true to pay the repo walk cost and get an exact count/details for files not yet in the index. Trust/stability is not freshness; this tool answers whether the index snapshot is current.
Inputs: projectId, verbosity, projectRef, includeUnindexed. Returns: watch, _hints, toolName, freshness, +9 more
project_open_loopsproject_open_loopsARead-onlyIdempotent
Reef 8 open-loop view: list unresolved active findings, stale or unknown facts, and stale or failed diagnostic runs so an agent can see what still needs attention without rerunning broad checks.
Inputs: limit, filePath, projectId, projectRef, +2 more. Returns: loops, _hints, summary, filePath, +5 more
recall_answersrecall_answersARead-onlyIdempotent
Read-only session-recall tool: search prior project answer traces by text, query kind, support level, trust state, and ISO time window. Returns bounded answer summaries, pre-cap matchCount, truncation signal, and optional stored answer markdown. Does not run the answer loop or infer freshness.
Inputs: limit, query, since, until, +5 more. Returns: _hints, answers, toolName, warnings, +4 more
recall_tool_runsrecall_tool_runsARead-onlyIdempotent
Read-only session-recall tool: inspect prior project tool_runs by tool name, outcome, request id, and ISO time window. Returns bounded summaries with runId plus requestId when available; use requestId as agent_feedback.referencedRequestId. Default limit is 50, max 500. Persisted payloads are only included when includePayload is true.
Inputs: limit, since, until, outcome, +5 more. Returns: _hints, toolName, toolRuns, warnings, +4 more
reef_agent_statusreef_agent_statusARead-onlyIdempotent
Reef agent-loop status summary: report maintained known issues, changed files needing verification, stale diagnostic sources, schema freshness, watcher degradation, and background queue state. Never edits files.
Inputs: limit, projectId, focusFiles, projectRef, +1 more. Returns: state, _hints, schema, summary, +9 more
reef_askreef_askARead-onlyIdempotent
Primary Reef query engine facade: ask one project question and receive a compiled codebase plus database evidence packet with compact answer sections, ranked context, findings, risks, open loops, diagnostic freshness, a normalized evidence graph, and suggested next actions. Internally handles bounded quoted literal checks, materialized DB inventories, specific table/column/RLS/index/FK/trigger questions, where-used/impact questions, and verification-gate questions with typed summaries. Raw evidence defaults to compact: per-section item caps plus omission of the raw evidence-graph nodes/edges and heavy per-item provenance/diagnostic detail (node/edge and section totals are still reported, and long finding messages are clipped). Pass evidenceMode="full" for the uncapped raw evidence and full graph, or includeTrace=true for the decision trace and engine steps. Read-only; uses internal Reef/context modules instead of model-orchestrated MCP tool chains.
Inputs: question (required), mode, projectId, focusFiles, +18 more. Returns: _hints, answer, limits, evidence, +8 more
reef_diff_impactreef_diff_impactARead-onlyIdempotent
Reef mid-edit impact composer: for changed working-tree files, return affected import callers, active caller findings that may need re-checking, and conventions the diff or impacted callers may violate. Reads existing working_tree_overlay facts and import graph state; it does not refresh or mutate Reef.
Inputs: filePaths (required), depth, projectId, projectRef, +4 more. Returns: _hints, filters, summary, toolName, +8 more
reef_impactreef_impactARead-onlyIdempotent
Thin Reef impact query: for changed files, return affected import callers, invalidated findings, and convention risks through the primary Reef surface. Equivalent calculation to reef_diff_impact with a smaller model-facing name.
Inputs: filePaths (required), depth, projectId, projectRef, +4 more. Returns: _hints, filters, summary, toolName, +8 more
reef_inspectreef_inspectARead-onlyIdempotent
Reef 7 inspection view: explain one file or subject fingerprint by returning its durable facts, findings, and relevant diagnostic runs. Use after reef_scout when the model needs the evidence trail and freshness details before editing.
Inputs: limit, filePath, projectId, projectRef, +1 more. Returns: facts, _hints, summary, filePath, +8 more
reef_instructionsreef_instructionsARead-onlyIdempotent
Reef read tool for scoped project instructions: load .mako/instructions.md and applicable AGENTS.md files for requested project-relative files, returning structured instruction items plus derived Reef fact shapes. Reads only inside the project root and does not persist facts.
Inputs: files, projectId, projectRef, includeDerivedFacts. Returns: files, _hints, summary, toolName, +5 more
reef_known_issuesreef_known_issuesCRead-onlyIdempotent
Reef warm issue query: answer "have any errors?" from maintained Reef findings and diagnostic readiness state without launching broad lint or typecheck commands.
Inputs: files, limit, sources, projectId, +4 more. Returns: _hints, issues, summary, toolName, +5 more
reef_learning_reviewreef_learning_reviewARead-onlyIdempotent
Suggestion-only Reef learning review: propose rule-pack, sentinel, instruction, convention, conjecture, or session-recall candidates from resolved findings, repeated rule history, recent tool runs, and agent feedback. Never writes durable knowledge; acceptance must happen through an explicit separate write path.
Inputs: mode, limit, since, projectId, +5 more. Returns: mode, _hints, summary, toolName, +6 more
reef_overlay_diffreef_overlay_diffARead-onlyIdempotent
Reef read tool for overlay comparison: diff durable facts between two overlays, defaulting indexed vs working_tree, with optional filePath/kind/source filters. Use after working_tree_overlay or before edits to see exactly what Reef believes changed without rerunning diagnostics.
Inputs: kind, limit, source, filePath, +6 more. Returns: _hints, entries, summary, toolName, +5 more
reef_scoutreef_scoutARead-onlyIdempotent
Reef 7 model-facing scout view: turn a messy request into intent-weighted, explainable candidates from durable Reef facts, findings, rules, and diagnostic runs. App-flow queries prefer files/routes/findings; RLS/schema queries prefer database evidence. This is a context scout, not an editing agent; use normal harness read/search after the packet.
Inputs: query (required), limit, projectId, focusFiles, +2 more. Returns: facts, query, _hints, findings, +7 more
reef_statusreef_statusBRead-onlyIdempotent
Thin Reef status query: report maintained known issues, changed files needing verification, stale diagnostic sources, schema freshness, watcher degradation, and background queue state through the primary Reef surface.
Inputs: limit, projectId, focusFiles, projectRef, +1 more. Returns: state, _hints, schema, summary, +9 more
reef_verifyreef_verifyCRead-onlyIdempotent
Thin Reef verification query: combine diagnostic freshness, changed files, watcher state, recent diagnostic runs, and unresolved open loops into one completion gate. Read-only and does not run diagnostics.
Inputs: files, limit, sources, projectId, +5 more. Returns: _hints, status, summary, toolName, +7 more
reef_where_usedreef_where_usedARead-onlyIdempotent
Reef maintained structural query: answer where a symbol, component, route, file, or indexed pattern is defined and used from maintained symbols/imports/routes first, then supplement symbol/component answers with indexed identifier-text references and related durable findings. Does not run grep; coverage explains import-graph/text/finding limits and returns fallback tool args when maintained state has no answer.
Inputs: query (required), limit, projectId, projectRef, +2 more. Returns: query, _hints, usages, coverage, +10 more
repo_maprepo_mapARead-onlyIdempotent
Code-intelligence tool for repo orientation: emit a token-budgeted aider-style outline of the indexed project (ranked files + key symbols) as first-turn context for agents meeting an unfamiliar codebase. Ranking uses import-graph PageRank, personalized bidirectionally around focusFiles, focusRoutes, focusSymbols, or focusDatabaseObjects when supplied so nearby dependencies and dependents surface first. Symbol selection prefers exported declarations. Read-only; default budget 1024 tokens (char/4 approximation), cap 16384.
Inputs: maxFiles, pathGlob, projectId, focusFiles, +6 more. Returns: files, _hints, rendered, toolName, +8 more
review_bundle_artifactreview_bundle_artifactCRead-onlyIdempotent
Artifact tool for reviewer-facing change inspection: compose implementation brief, change plan, optional flow map, and optional tenant audit into one typed review bundle. Basis is stable — identical inputs produce the same artifactId, so callers may dedupe on it.
Inputs: queryKind (required), queryText (required), startEntity (required), targetEntity (required), +12 more. Returns: _hints, result, exported, toolName, +1 more
route_contextroute_contextARead-onlyIdempotent
Neighborhood composer for one route: combines route_trace route resolution, file_health handler summary, imports_deps direct imports/dependents, schema_usage touches, trace_rpc downstream table edges, and db_rls policy surfaces.
Inputs: route (required), projectId, projectRef, maxPerSection. Returns: route, trust, _hints, toolName, +12 more
route_traceroute_traceBRead-onlyIdempotent
Answer tool for route questions: trace a route to the indexed handler, matching files, and nearby evidence.
Inputs: route (required), projectId, projectRef. Returns: _hints, result, toolName, projectId
rpc_neighborhoodrpc_neighborhoodBRead-onlyIdempotent
Neighborhood composer for one RPC: combines db_rpc-equivalent schema snapshot signature/body, schema_usage app-code callers, trace_rpc function-to-table refs, and db_rls policies on touched tables into one bounded bundle.
Inputs: rpcName (required), argTypes, projectId, projectRef, +2 more. Returns: rpc, trust, _hints, callers, +10 more
rule_memoryrule_memoryARead-onlyIdempotent
Reef 9 rule-memory view: aggregate Reef rule descriptors and finding history so agents can see which rules are active, acknowledged, resolved, or suppressed before making changes.
Inputs: limit, projectId, projectRef, sourceNamespace. Returns: _hints, entries, toolName, warnings, +3 more
rule_pack_validaterule_pack_validateARead-onlyIdempotent
Read-only Reef rule-pack authoring tool: validate .mako/rules YAML packs, including canonicalHelper cross-file declarations; return pack/rule counts, schema errors, and optional ReefRuleDescriptor previews without running diagnostics or writing Reef state.
Inputs: projectId, projectRef, includeDescriptors. Returns: packs, rules, _hints, summary, +4 more
runtime_telemetry_reportruntime_telemetry_reportARead-onlyIdempotent
Read-only inspection over mako_usefulness_events (Phase 8.1): return aggregate counts (by decisionKind, by family, by grade) and a bounded list of recent events, filterable by decisionKind, family, requestId, and ISO time window. Use to verify usefulness capture is flowing or to triage why a decision site emitted the grade it did. Never writes to the event table.
Inputs: limit, since, until, family, +4 more. Returns: _hints, events, byGrade, byFamily, +7 more
schema_usageschema_usageARead-onlyIdempotent
Answer tool for schema questions: find where an indexed schema object is defined and directly referenced in app code. RPC-mediated touches are intentionally excluded; use trace_rpc, route_context, table_neighborhood, or flow_map for transitive schema paths.
Inputs: object (required), schema, projectId, projectRef. Returns: _hints, result, toolName, projectId, +2 more
session_handoffsession_handoffCRead-onlyIdempotent
Operator tool for derived project-state handoff: summarize recent answer traces, active unresolved focus, and recorded follow-up momentum.
Inputs: limit, projectId, projectRef. Returns: _hints, result, toolName, projectId
shell_runADestructive
Run a shell command with arguments as a list (never concatenated). cwd is locked to the project root or a subdirectory; env keys must be allowlisted.
Inputs: args (required), command (required), cwd, env, +1 more. Returns: ok, error, _hints, requiresHarnessSession
suggestsuggestBRead-onlyIdempotent
Workflow tool for tool-chain recommendations: choose one canonical workflow or a short ordered sequence without executing hidden planner logic.
Inputs: question (required), maxSteps, direction, projectId, +5 more. Returns: _hints, result, toolName, projectId
symbols_ofsymbols_ofBRead-onlyIdempotent
Symbols tool for declarations: list the indexed symbols declared in a file.
Inputs: file (required), projectId, projectRef. Returns: file, _hints, symbols, toolName, +4 more
table_neighborhoodtable_neighborhoodARead-onlyIdempotent
Neighborhood composer for one table: combines db_table_schema/db_rls-equivalent schema snapshot data, schema_usage read/write sites, trace_table RPC-to-table edges, and route_trace handler routes into one bounded typed bundle.
Inputs: tableName (required), projectId, projectRef, schemaName, +1 more. Returns: rls, reads, table, trust, +11 more
task_preflight_artifacttask_preflight_artifactARead-onlyIdempotent
Artifact tool for start-of-work preparation: compose implementation brief, verification plan, change plan, and flow map into one typed preflight artifact. Basis is stable — identical inputs produce the same artifactId, so callers may dedupe on it.
Inputs: queryKind (required), queryText (required), startEntity (required), targetEntity (required), +8 more. Returns: _hints, result, exported, toolName, +1 more
tenant_leak_audittenant_leak_auditBRead-onlyIdempotent
Operator tool for tenant-boundary review: audit tenant-keyed tables, RLS posture, and RPC/code touch points without making generic security-score claims.
Inputs: acknowledgeAdvisory (required), freshen, projectId, projectRef, +2 more. Returns: _hints, result, toolName, projectId
tool_batchtool_batchARead-only
Read-only batching wrapper for independent Mako lookups. The input schema only accepts batchable read-only tools; runtime also rejects mutation tools and recursive tool_batch calls defensively. With continueOnError=true, operations run with bounded concurrency while results preserve input order; tune maxConcurrency to trade latency against shared-store pressure. continueOnError=false keeps fail-fast sequential execution. Returns labeled sub-results with per-op duration plus summary latency metadata including totalOpDurationMs and slowestOp. Use verbosity: "compact" or per-op resultMode: "summary" to return compact summaries instead of full payloads; per-op resultMode: "full" keeps the full output. Each sub-op's projectId is overridden with the parent batch's resolved project, so all ops run against one project. Use to reduce round-trips after a context_packet recommends several expansions.
Inputs: ops (required), maxOps, projectId, verbosity, +3 more. Returns: _hints, results, summary, toolName, +3 more
tool_searchtool_searchARead-onlyIdempotent
Search the MCP-visible tool catalog, including deferred and blocked tools. Use when you are unsure which tool fits a task or why a tool is unavailable over MCP.
Inputs: query (required), limit. Returns: count, query, _hints, results
trace_edgetrace_edgeARead-onlyIdempotent
Trace a handler / edge function: its own route, app-code callers (ast-grep on fetch('/functions/v1/$NAME') and supabase.functions.invoke('$NAME')), tables and RPCs the handler touches, and DB triggers whose body references the name. Snapshot-strict.
Inputs: name (required), projectId, projectRef. Returns: _hints, result, toolName, projectId
trace_errortrace_errorARead-onlyIdempotent
Trace an error term across throw sites (ast-grep throw new Error($MSG), throw new $ERR($MSG)), catch handlers (ast-grep try/catch), and PL/pgSQL bodies that reference the term. FTS narrows the ast-grep pass. Snapshot-strict.
Inputs: term (required), projectId, projectRef. Returns: _hints, result, toolName, projectId
trace_filetrace_fileARead-onlyIdempotent
Trace a file end-to-end from the snapshot: declared symbols, outbound imports, inbound dependents, routes contributed, and related evidence. Read-only.
Inputs: file (required), projectId, projectRef. Returns: _hints, result, toolName, projectId
trace_rpctrace_rpcARead-onlyIdempotent
Trace an RPC end-to-end: the RPC definition (searchSchemaObjects filtered to rpc), other PL/pgSQL bodies whose body text references it (searchSchemaBodies), overload-aware table refs (listFunctionTableRefs), and app-code .rpc('$FN') call sites (FTS-retrieval + ast-grep proof). Snapshot-strict.
Inputs: name (required), schema, argTypes, projectId, +1 more. Returns: _hints, result, toolName, projectId
trace_tabletrace_tableARead-onlyIdempotent
Trace a table end-to-end: columns, indexes, foreign keys, RLS, triggers (via getSchemaTableSnapshot), schema-scoped RPC → table edges (via listFunctionTableRefs), and app-code .from('$TABLE') call sites (FTS-retrieval + ast-grep proof). Snapshot-strict.
Inputs: table (required), schema, projectId, projectRef. Returns: _hints, result, toolName, projectId
typescript_diagnosticstypescript_diagnosticsA
Explicit Reef ingestion tool for semantic TypeScript compiler diagnostics: read tsconfig, run the TypeScript program with no emit, persist working-tree ProjectFinding rows under source typescript, and return bounded findings. This may run a broad typecheck; use diagnostic_refresh with typescript_syntax for syntax-only refresh.
Inputs: files, projectId, projectRef, maxFindings, +1 more. Returns: _hints, status, findings, toolName, +11 more
verification_bundle_artifactverification_bundle_artifactARead-onlyIdempotent
Artifact tool for safe completion and verification: compose verification plan, project-intelligence signals, and optional tenant audit into one typed verification bundle. Basis is stable by default; artifactId shifts across calls only when includeSessionHandoff or includeIssuesNext is set, because those inputs are session-scoped. Dedupe is safe when both flags are off.
Inputs: queryKind (required), queryText (required), export, traceId, +9 more. Returns: _hints, result, exported, toolName, +1 more
verification_stateverification_stateARead-onlyIdempotent
Reef 8 verification state view: summarize whether cached diagnostic runs still cover the current working-tree overlay, including file-scoped recent runs, watcher diagnostic state, files modified after successful checks, and suggested verification actions. With files, runs only count when project-wide or scoped to those files.
Inputs: files, limit, sources, projectId, +2 more. Returns: _hints, status, sources, watcher, +8 more
workflow_packetworkflow_packetCRead-onlyIdempotent
Workflow tool: generate a typed workflow packet from a project-scoped query answer and expose watch/surface metadata.
Inputs: family (required), queryKind (required), queryText (required), scope, +9 more. Returns: _hints, result, toolName, projectId
working_tree_overlayworking_tree_overlayB
Advisory Reef mutation: snapshot working-tree file facts for explicit files, watcher-dirty paths, or non-fresh indexed paths. Persists file_snapshot facts with size, mtime, line count, sha256, or deleted state; it does not reparse AST, imports, routes, or schema.
Inputs: files, maxFiles, projectId, projectRef, +1 more. Returns: facts, _hints, toolName, warnings, +5 more
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 105 tools
With 105 tools, several are near-duplicates: reef_status and reef_agent_status describe essentially the same status report, and reef_impact is explicitly an equivalent of reef_diff_impact. Large clusters around status, verification, diagnostics, and querying blur boundaries, so an agent will frequently struggle to pick the right tool.
Most names are readable snake_case with useful prefixes like trace_*, db_*, and file_*, but the conventions are mixed: verb_noun (trace_table), noun_noun (schema_usage), and bare verbs (ask, suggest, investigate) all appear. The inconsistent word order and singular/plural choices across families prevent a fully predictable pattern.
105 tools is an extreme count for any MCP server, far beyond the 50+ threshold for a 1. The breadth does not justify the size because many tools are near-duplicates or slight variants rather than distinct capabilities.
Coverage is broad across code search, tracing, database inspection, diagnostics, findings, and session recall, but there is no simple file read tool despite full file mutation tools, and the four artifact composers have no artifact read/list counterpart. Shell_run and live_text_search can work around some gaps, but the lifecycle is not complete.
Maintenance
Related MCP Connectors
Codebase intelligence for AI coding agents: graph, git history, docs, decisions, code health.
Code intelligence for coding agents: semantic, AST, graph, and full-text search. 279+ languages.
Team memory for AI coding agents and code reviews, reviewed by humans and served over MCP.
Shared context for your codebase: a hosted code knowledge graph over MCP. Free plan, OAuth sign-up.
Related MCP Servers
- AlicenseBqualityDmaintenanceLocal-first codebase context engine that parses code into a ranked dependency graph and serves it to AI tools via MCP for deep structural understanding.518 npm2MIT
- AlicenseNot gradedqualityDmaintenanceA local-first codebase intelligence layer for AI coding agents, providing a persistent, queryable model of a repository via an MCP server and CLI to enable structure queries instead of reading many files.1Apache 2.0
- FlicenseNot gradedqualityCmaintenanceProvides a local-first code indexing and search engine for coding agents via MCP, enabling precise codebase queries, symbol lookup, and freshness-aware retrieval.-
- AlicenseNot gradedqualityBmaintenanceLocal-first code intelligence MCP server that enables coding agents to search code, inspect structure, read exact ranges, and explore Git history with explicit token budgets.171Apache 2.0