FusionAL-Recall
MCP server for semantic search over a solved-issues engineering knowledge base. Built on FastMCP with sqlite-vec and sentence-transformers; exposes recall, remember, list_recent, verify, and get tools so agents can retrieve and log fixes.
README
FusionAL Recall — Semantic Registry Search MCP
Recall wraps the FusionAL Solved-Issues Registry with semantic search, three-tier access control, and auto-migration.
Query your team's solved problems via MCP tools: recall() for semantic search, remember() to log new entries, list_recent() for recent issues, verify() to validate entry integrity.
Features
- Semantic Search: Query by problem description, not keyword. "claude desktop timeout" returns SI-001 even if you don't say "timeout" or "claude" exactly.
- Automatic Migration: Parse existing SOLVED-ISSUES.md markdown on first run; auto-assign SI-XXX IDs; populate SQLite.
- Three-Tier Access:
personal(agent-only),project(team-scoped),public(shared community). - Vector Storage: sentence-transformers + sqlite-vec for fast CPU-friendly embeddings.
- MCP Tools:
recall(query, tier, limit),remember(symptoms, root_cause, fix, source, tags),list_recent(n),verify(si_id),get(si_id).
Architecture
SOLVED-ISSUES.md (markdown source)
↓
[Migrate on startup]
↓
SQLite + sqlite-vec (embeddings index)
↓
FastMCP Server (port 8107)
↓
[Tools]
SQL Schema
CREATE TABLE issues (
si_id TEXT PRIMARY KEY, -- SI-001, SI-002, etc.
title TEXT NOT NULL, -- one-line summary
symptoms TEXT NOT NULL,
root_cause TEXT NOT NULL,
fix TEXT NOT NULL,
source TEXT, -- session/task context
tags TEXT, -- comma-separated
verified_at TEXT, -- YYYY-MM
created_at TEXT NOT NULL, -- ISO 8601
tier TEXT DEFAULT 'personal', -- personal, project, public
embedding BLOB NOT NULL -- float32 vec (float-serialized)
);
Usage
Start the Server
python -m recall.server
# Listens on 0.0.0.0:8107
On first run, the server:
- Reads SOLVED-ISSUES.md (path from SOLVED_ISSUES_PATH env var)
- Parses SI-XXX markdown blocks
- Generates embeddings for symptoms + root_cause + fix (combined text)
- Inserts into SQLite + sqlite-vec index
- Auto-assigns next SI-ID based on highest existing ID
MCP Tools
recall(query: str, tier: str = "personal", limit: int = 5) → List[Issue]
Semantic search. Returns issues matching the query, ranked by embedding similarity.
# Query: "claude desktop timeout"
# Returns: SI-001 (Claude Desktop server timeout above 8 servers)
results = await client.call_tool(
name="recall",
arguments={"query": "claude desktop timeout", "tier": "personal", "limit": 5}
)
remember(symptoms: str, root_cause: str, fix: str, source: str, tags: str) → Dict
Log a new issue. Returns the assigned SI-ID (e.g., SI-012).
result = await client.call_tool(
name="remember",
arguments={
"symptoms": "My symptoms here",
"root_cause": "Root cause",
"fix": "Steps to fix",
"source": "session context",
"tags": "tag1,tag2"
}
)
# Returns: {"si_id": "SI-012", "title": "...", "created_at": "..."}
list_recent(n: int = 10) → List[Issue]
Return the N most recently added issues.
recent = await client.call_tool(
name="list_recent",
arguments={"n": 10}
)
verify(si_id: str) → Dict
Check if an SI entry exists and is valid.
valid = await client.call_tool(
name="verify",
arguments={"si_id": "SI-001"}
)
get(si_id: str) → Dict
Retrieve full entry by SI-ID.
entry = await client.call_tool(
name="get",
arguments={"si_id": "SI-001"}
)
Deployment
Docker
docker build -t fusional-recall:latest .
docker run -p 8107:8107 \
-e SOLVED_ISSUES_PATH=/mnt/kb/05-RECALL/SOLVED-ISSUES.md \
-v /path/to/kb:/mnt/kb \
fusional-recall:latest
Docker Compose
docker-compose up -d
# Recalls listens on localhost:8107
# recall.db volume persists embeddings across restarts
Testing
pytest tests/ -v
Tests validate:
- Migration: SOLVED-ISSUES.md parses correctly; all entries load into SQLite
- Recall: Semantic search returns SI-001 for query "claude desktop timeout"
- Remember: New entry is assigned SI-012, written to DB, queryable immediately
- Verify: SI-001 exists and validates; SI-999 does not
- Tier filtering: personal/project/public tiers are enforced
Environment Variables
| Variable | Default | Purpose |
|---|---|---|
RECALL_HOST |
0.0.0.0 |
MCP server bind address |
RECALL_PORT |
8107 |
MCP server port |
RECALL_DB_PATH |
./recall.db |
SQLite database file |
EMBEDDING_MODEL |
all-MiniLM-L6-v2 |
sentence-transformers model |
DEFAULT_TIER |
personal |
Default access tier |
SOLVED_ISSUES_PATH |
../fusional-knowledge-base/05-RECALL/SOLVED-ISSUES.md |
Path to registry markdown |
License
MIT. See LICENSE.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.