abhigyan-shekhar avatar

Waggle Mcp

abhigyan-shekhar/waggle-mcp

Waggle gives Claude and other MCP clients a persistent graph memory that survives context windows. It stores decisions, their reasons, and contradictions as typed nodes and edges in SQLite or Neo4j, then retrieves relevant subgraphs on demand. You get three core tools: query_graph for retrieval, observe_conversation to persist turns, and prime_context to hydrate sessions. Setup is one command (waggle-mcp setup --yes) that auto-configures Claude Code, Cursor, and others. Includes local sentence-transformers embeddings with a deterministic fallback, so no API keys required. Ships with Graph Studio for inspection and a demo mode that runs sample queries against a preloaded graph. Built for agents that need to remember what they decided three sessions ago and why.

GitHub stars
16
tools
7
transport
STDIO
env vars
5
last updated
Apr 21, 2026

Tools

7 tools

Public tool metadata for what this MCP can expose to an agent.

searchFind A2A agents by capability or use case using semantic search.10 params

Find A2A agents by capability or use case using semantic search.

Parameters* required
autharray
Filter by auth type (e.g. ["public"])
capabilitiesarray
Required capabilities (e.g. ["streaming"])
healthy_onlyboolean
Only return currently online agents
input_modesarray
Filter by input MIME types
limitinteger
Max results (1-20, default 10)
offsetinteger
Pagination offset (default 0)
output_modesarray
Filter by output MIME types
providerstring
Fuzzy provider name search
querystring
Natural language search query
verified_onlyboolean
Only return verified agents
invokeSend a message to an A2A agent directly (pass-through proxy). Provide either url_hash (from search) or agent_url (direct URL).7 params

Send a message to an A2A agent directly (pass-through proxy). Provide either url_hash (from search) or agent_url (direct URL).

Parameters* required
agent_urlstring
Direct agent URL (required if url_hash omitted)
context_idstring
Context ID for agents with server-side history
conversation_idstring
Conversation ID for follow-ups and rating linkage
messagestring
Message to send (max 2000 chars)
protocol_versionstring
A2A protocol version (auto-detected if omitted)
task_idstring
Task ID for follow-up or async polling
url_hashstring
Agent URL hash from search results (required if agent_url omitted)
pollCheck the status of an async A2A task.5 params

Check the status of an async A2A task.

Parameters* required
agent_urlstring
The agent's URL
context_idstring
Context ID for server-side history
conversation_idstring
Conversation ID for tracking
protocol_versionstring
A2A protocol version
task_idstring
Task ID returned by the agent
infoGet full details about a specific agent — skills, health, ratings, and capabilities.1 params

Get full details about a specific agent — skills, health, ratings, and capabilities.

Parameters* required
url_hashstring
Agent URL hash
registerSubmit a new A2A agent URL for indexing in Waggle.1 params

Submit a new A2A agent URL for indexing in Waggle.

Parameters* required
urlstring
Agent URL to register
check_usageView your current API quota, daily usage, and rate limit status.

View your current API quota, daily usage, and rate limit status.

No parameter schema in public metadata yet.

rate_transactionRate an agent interaction to improve ranking quality.6 params

Rate an agent interaction to improve ranking quality.

Parameters* required
conversation_idstring
Conversation ID from invoke/poll
notesstring
Optional feedback (max 500 chars)
qualityinteger
Quality rating 1-5
rating_idstring
Optional idempotency key
successboolean
Whether the interaction was successful
url_hashstring
Agent URL hash

Sponsors

Configuration

WAGGLE_TRANSPORTdefault: stdio

Transport mode for the MCP server.

WAGGLE_BACKENDdefault: sqlite

Backend database type: sqlite for local use or neo4j for service deployments.

WAGGLE_DB_PATHdefault: ~/.waggle/memory.db

Path to the SQLite memory database when WAGGLE_BACKEND is sqlite.

WAGGLE_DEFAULT_TENANT_IDdefault: local-default

Default tenant ID for local or shared memory isolation.

WAGGLE_MODELdefault: all-MiniLM-L6-v2

Sentence-transformers model used for local embeddings.

Readme

Waggle

Project memory for humans and AI agents.
Keep decisions, context, and the reasons behind them across conversations.

PyPI Python 3.11+ MCP compatible Apache-2.0

Install Waggle · Try the browser workspace · Documentation


Keep the context, not just the conversation


View the full README on GitHub

Details

Repository
github.com/abhigyan-shekhar/waggle-mcp
Category
AI & LLM Tools
Package
waggle-mcp
Transport
STDIO
Registry
MCP registry, active

Related AI & LLM Tools MCP servers

View all
1unblocked avatarUnblockedunblocked/skillsUnblocked is the context layer for agentic software development.0stars
2ejentum avatarEjentum Mcpejentum/ejentum-mcpReasoning, code, anti-deception, memory harness MCP tools. Stdio or HTTPS api.ejentum.com/mcp16stars
3gambletan avatarCortex Memory Enginegambletan/cortexLocal-first AI memory engine — 4-tier memory, people graph, Bayesian beliefs. Encrypted, 62µs.16stars
4multiplex-ai avatarMuggle Worksmultiplex-ai/muggle-ai-worksAcceptance test agent for coding agents: real-browser tests, with screenshots on the PR.16stars
5camgitt avatarMemoircamgitt/memoirPersistent memory for AI coding tools via MCP. Remembers across sessions and machines.15stars
6buywhere avatarBuywhere Mcpbuywhere/buywhere-mcpAgent-native product catalog: 300M+ products, 150,000+ stores, deliver_to ranking.15stars