Skip to main content
Glama
ArkTechNWA

Federated Search

by ArkTechNWA
OverviewSchemaRelated ServersScoreDiscussions
Python
Remote

Federated Search

One query, all knowledge. A federation MCP server that sits in front of multiple memory backends and presents a unified search surface to AI agents.

What It Does

Instead of an agent juggling 3-4 MCP connections and deciding which backend to query, federation handles it:

Agent → fed_search("Kronos") → Federation
                                   ├→ Knowledge Graph (curated entities)
                                   ├→ Flex (session history)
                                   └→ SearXNG (web, opt-in)
                                ← merged, ranked, one response

Results are priority-ordered by bank, relevance-ranked within each bank, and filtered for signal quality.

Related MCP server: Qdrant Neo4j Crawl4AI MCP Server

Tools

fed_search(query, db?, limit?, mode?, domain?)

Search across all subscribed memory banks.

Parameter

Default

Description

query

required

What to search for

db

all defaults

Bank ID or comma-separated IDs. "knowledge_graph", "flex,web"

limit

10

Max results. -1 for unlimited

mode

"broad"

"broad", "exact" (phrase match), or "semantic" (meaning-based)

domain

none

Pre-filter KG results to an index alias. "infrastructure", "api"

fed_banks()

Returns registered banks with priorities, descriptions, and health status.

Architecture

federation/
  server.py          # FastMCP server — tool definitions
  federation.py      # Core engine — fan-out, merge, rank
  config.py          # YAML config loader
  types.py           # FederatedResult envelope, BankConfig, SearchRequest
  filters.py         # Signal quality — confidence floor, adaptive count, dedup
  formatter.py       # Markdown output formatting
  plugins/
    base.py          # BankPlugin ABC
    kg.py            # Knowledge graph MCP plugin
    flex.py          # Flex session history plugin
    searxng.py       # SearXNG web search plugin

Plugin System

Each backend is a plugin that translates fed_search into native queries and packs results into a universal envelope:

class MyPlugin(BankPlugin):
    async def search(self, query, limit=10, mode="broad", domain=None):
        # Call your backend, return list[FederatedResult]

    async def health_check(self):
        # Return BankStatus.HEALTHY / DEGRADED / DOWN

Adding a new bank = write a plugin class + add a YAML config block. No core changes.

See skills/federation-plugin-dev.md for the full plugin development guide.

Signal Quality

  • Query validation — rejects empty, single-char, and stopword queries

  • Confidence floor — results below 0.25 relevance get cut

  • Adaptive count — when strong results exist, weak tail is trimmed with a note

  • Bank representation — each bank gets at least 1 result slot

  • Cross-bank annotation — flex chunks referencing KG entities get overlaps_with metadata

Config

config.yaml defines agents and their bank subscriptions:

agents:
  my_agent:
    port: 4001
    banks:
      - id: knowledge_graph
        type: kg
        label: "Curated Knowledge"
        description: "Agent-curated structured knowledge graph"
        priority: 1        # lower = results sort first
        default: true       # searched when no db= specified
        url: "http://127.0.0.1:3101/mcp"
        auth: "Bearer ${KG_AUTH_TOKEN}"
      - id: web
        type: searxng
        priority: 99
        default: false      # opt-in only
        url: "http://your-searxng:8080"

Copy config.yaml to config.local.yaml and fill in real values. The local config is gitignored.

Setup

python3 -m venv .venv
source .venv/bin/activate
pip install -e .

Usage

# stdio mode (for Claude Code MCP)
python -m federation.server --agent my_agent --config config.local.yaml

# HTTP mode
python -m federation.server --agent my_agent --config config.local.yaml --http --port 4001

Add to Claude Code

claude mcp add fed-search -s user -- \
  /path/to/.venv/bin/python -m federation.server \
  --agent my_agent --config /path/to/config.local.yaml

License

MIT

This server cannot be installed

F
license - not found
-
quality - not tested
B
maintenance

How are these scores calculated?

Maintenance

–Maintainers
–Response time
–Release cycle
–Releases (12mo)
Commit activity

Resources

Unclaimed servers have limited discoverability.

Looking for Admin?

If you are the server author, to access and configure the admin panel.

Related MCP Servers

  • pce-memory

    CAPHTECH
    A
    license
    -
    quality
    C
    maintenance
    MCP server for context-aware memory and retrieval with hybrid search, dual-phase memory, and boundary-first security.
    3
    MIT

  • knowledgeplane

    camplight
    F
    license
    -
    quality
    C
    maintenance
    MCP server that gives AI agents and teams persistent, shared memory using a knowledge graph with vector embeddings, automatic consolidation of related facts, and hybrid search.
    3

  • jade-memory

    jjjona
    F
    license
    -
    quality
    D
    maintenance
    Two MCP servers (knowledge base and journal) enabling AI agents to persistently store, search, and recall knowledge and reflections with multilingual semantic search.

View all related MCP servers

Related MCP Connectors

  • XMemo

    User-owned memory for AI agents, Copilot, Claude, IDEs, CLIs, and chat apps over remote MCP.

  • mcp

    Your memory, everywhere AI goes. Build knowledge once, access it via MCP anywhere.

  • qsearch

    Multi-engine search for AI agents. Trust scoring, local corpus, MCP-native. Self-hostable, BYOK.

View all MCP Connectors

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/ArkTechNWA/federated-search'

If you have feedback or need assistance with the MCP directory API, please join our Discord server