Atlas / MCP servers / ruvnet / AgentDB

AgentDBBLOCK

Name: AgentDB
Author: ruvnet

mcp/ruvnet/agentdb

MCP server by ruvnet · agent memory · Audited 2026-10-07

⚑ Report / re-scan Source ↗

Vector memory that gets smarter every time your agent uses it.

Verdict

BLOCK

Grade

Trust score

26 /100

Exposed tools

76 62r · 11w · 3d

Transport

stdio

License

MIT

Stars

Overview

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

[](https://www.npmjs.com/package/agentdb) [](https://www.npmjs.com/package/agentdb) [](LICENSE) [](https://www.typescriptlang.org/)

[](https://github.com/ruvnet/ruvector) [](#-self-learning-loop) [](#-mcp-integration) [](https://github.com/ruvnet/agentdb)

Vector memory that gets smarter every time your agent uses it.

A single-file cognitive container — vectors, indexes, learning state, and a cryptographic audit trail in one .rvf. Self-learning search improves up to 36% from feedback alone, with no manual tuning. Runs in Node, the browser, edge runtimes, and offline.

Why AgentDB?

Most vector databases store embeddings and call it done. AgentDB watches which results your agent actually used, learns from that signal, and ranks the next query better. The bandit underneath also picks the right RL algorithm, the right compression tier, and the right pattern weighting on its own — so the database itself gets sharper while you focus on the agent.

The name: a database that thinks like an agent — episodic memory, skill library, causal r

Read from source at commit 85bed1bd0078OBSERVED · 2026-10-07

Connect

Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control. Replace the environment placeholders with a token scoped to the least it needs.

claude-code

claude mcp add vite_react_shadcn_ts --env AGENTDB_DEV_BYPASS_SECRET=${AGENTDB_DEV_BYPASS_SECRET} --env ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY} --env GOOGLE_GEMINI_API_KEY=${GOOGLE_GEMINI_API_KEY} --env HF_TOKEN=${HF_TOKEN} -- npx -y [email protected]

claude-desktop

{
  "mcpServers": {
    "vite_react_shadcn_ts": {
      "command": "npx",
      "args": [
        "-y",
        "[email protected]"
      ],
      "env": {
        "AGENTDB_DEV_BYPASS_SECRET": "${AGENTDB_DEV_BYPASS_SECRET}",
        "ANTHROPIC_API_KEY": "${ANTHROPIC_API_KEY}",
        "GOOGLE_GEMINI_API_KEY": "${GOOGLE_GEMINI_API_KEY}",
        "HF_TOKEN": "${HF_TOKEN}"
      }
    }
  }
}

Exposed tools (76)

62 read · 11 write · 3 destructive. Blast radius: 3 tools can delete or overwrite — an agent that can be talked into calling a tool can be talked into calling this one.

Tool	Risk	Description
`AgentDB`	read	Ultra-fast vector database for AI agents with WebAssembly acceleration, adaptive learning, and 100% client-side operation.
`Basketball`	read	Official size basketball for indoor/outdoor use
`advanced_auth`	read	Advanced authentication with MFA
`agentdb_attention_benchmark`	read	Benchmark attention mechanism performance
`agentdb_attention_compute`	read	Compute attention mechanism for query-key-value triplets
`agentdb_attention_configure`	read	Configure attention mechanism parameters
`agentdb_attention_metrics`	read	Get attention mechanism usage metrics and statistics
`agentdb_clear_cache`	destructive	Clear query cache to refresh statistics and search results
`agentdb_delete`	destructive	Delete vector(s) from AgentDB by ID or filters. Supports single ID deletion or bulk deletion with conditions.
`agentdb_delete_batch`	destructive	Delete many episode/pattern rows by ID list in a single transaction. Faster and safer than looping
`agentdb_init`	read	Initialize AgentDB database with schema and optimizations. Creates all required tables for vector storage, causal memory, skills, and provenance tracking.
`agentdb_insert`	write	Insert a single vector with metadata into AgentDB. Automatically generates embeddings for the provided text.
`agentdb_insert_batch`	write	Batch insert multiple vectors efficiently using transactions and parallel embedding generation. Optimized for large datasets.
`agentdb_pattern_search`	read	Search patterns with taskEmbedding, k, threshold, and filters
`agentdb_pattern_stats`	read	Get pattern statistics including total patterns, success rates, and top task types
`agentdb_pattern_store`	read	Store reasoning pattern with embedding, taskType, approach, and successRate
`agentdb_pattern_store_batch`	read	Batch store multiple reasoning patterns efficiently using transactions and parallel embedding generation. 4x faster than sequential agentdb_pattern_store calls. 🔄 PARALLEL-SAFE: Can be used alongside other batch operations.
`agentdb_search`	read	Semantic k-NN vector search using cosine similarity. Returns the most relevant results ranked by similarity score.
`agentdb_stats`	read	Get comprehensive database statistics including table counts, storage usage, and performance metrics
`api_request`	read	Make authenticated API requests
`auth_skill`	read	Authentication implementation
`authentication`	read	User authentication with JWT
`basic_auth`	read	Basic authentication
`cache_manager`	read	Redis-based caching with TTL
`caching-skill`	read	Implement caching strategies
`causal_add_edge`	write	Add a causal relationship between actions and outcomes
`causal_query`	read	Query causal effects to understand what actions cause what outcomes
`causal_traverse`	read	Walk the causal graph between two memories. Returns the chain of causal links (with confidence and uplift) up to max_depth hops. Use this to answer
`complex_skill`	read	Complex skill with nested types
`data_validation`	read	Input data validation
`database_query`	write	Execute SQL queries safely
`database_query_optimizer`	read	Optimize database queries with batch loading
`db_stats`	read	Get database statistics showing record counts
`error_handler`	read	Comprehensive error handling middleware
`experience_record`	read	Record tool execution as experience for reinforcement learning and experience replay
`extracted-pattern`	read	Pattern from learning
`fast-skill`	read	Fast execution skill
`fibonacci`	read	Calculate fibonacci numbers
`hash_password`	read	Secure password hashing with bcrypt
`high-success`	read	High success rate skill
`jwt_auth`	read	JWT authentication with tokens
`jwt_authentication`	read	Generate and verify JWT tokens
`learner_discover`	read	Automatically discover causal patterns from episode history
`learning_end_session`	write	End an active learning session and save the final trained policy to the database.
`learning_explain`	read	Explain action recommendations with confidence scores and supporting evidence from past experiences
`learning_feedback`	write	Submit feedback on action quality to train the RL policy. Feedback includes reward signal and outcome state.
`learning_metrics`	read	Get learning performance metrics including success rates, rewards, and policy improvement
`learning_predict`	read	Get AI-recommended action for a given state with confidence scores and alternative actions.
`learning_start_session`	write	Start a new reinforcement learning session with specified algorithm and configuration. Supports 9 RL algorithms: q-learning, sarsa, dqn, policy-gradient, actor-critic, ppo, decision-transformer, mcts, model-based.
`learning_train`	read	Train the RL policy using batch learning with collected experiences. Returns training metrics including loss, average reward, and convergence rate.
`learning_transfer`	write	Transfer learning between sessions or tasks, enabling knowledge reuse across different contexts
`metadata_test`	read	Test metadata persistence
`migration-skill`	read	Test skill for migration
`no_description`	read
`oauth2_login`	read	OAuth2 login flow with PKCE
`optimization-skill`	read	Optimize database queries
`recall_with_certificate`	read	Retrieve memories with causal utility scoring and provenance certificate
`reflexion_retrieve`	read	Retrieve relevant past episodes for learning from experience
`reflexion_store`	read	Store an episode with self-critique for reflexion-based learning
`reflexion_store_batch`	read	Batch store multiple episodes efficiently using transactions and parallel embedding generation. 3.3x faster than sequential reflexion_store calls (152 → 500 ops/sec). 🔄 PARALLEL-SAFE: Can be used alongside other batch operations.
`reward_signal`	read	Calculate reward signal for outcomes based on success, efficiency, and causal impact
`simple_skill`	read	A simple test skill
`skill-1`	read	First skill
`skill-2`	read	Second skill
`skill_create`	write	Create a reusable skill in the skill library
`skill_create_batch`	write	Batch create multiple skills efficiently using transactions and parallel embedding generation. 3x faster than sequential skill_create calls (304 → 900 ops/sec). 🔄 PARALLEL-SAFE: Can be used alongside other batch operations.
`skill_search`	read	Search for applicable skills by semantic similarity
`stats-skill`	read	Skill for stats testing
`tagged-pattern`	read	Pattern with tags
`test-pattern`	read	Test pattern storage
`test-skill`	read	Test skill
`test_skill`	read	Test skill
`test_update`	write	Test skill update
`type_test`	read	Type compatibility test
`unique-pattern`	read	First pattern
`validation_schema`	read	Request validation with Zod schemas

Trust audit

BLOCKgrade F · trust 26/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.

Layer	What it checks	Result
L0	Provenance & inventory	WARN
L1	Static analysis of the code	FAIL
L2	Instruction surface (what it tells the agent)	FAIL
L3	Class-specific surface	WARN
L4	Behavioural (sandbox)	SKIPPED