11mirror

11Mirror — Enterprise Knowledge Plugin

Enterprise knowledge layer for finance operations. Connects to company knowledge brain (people, playbooks, SOPs) and structured data graph (accounting, procurement, vendor payments, financial reports).

Built for Claude Cowork, also compatible with Claude Code.

Installation

Cowork

Install from claude.com/plugins or:

claude plugins add 11mirror

Claude Code

claude plugin install 11mirror

If you prefer to wire the gateway up by hand, copy examples/claude-code.mcp.json into your project root as .mcp.json and set ELEVEN_MIRROR_API_KEY in your environment.

OpenClaw

Loaded as a native OpenClaw bundle via openclaw.plugin.json. Configure the plugin with:

Key	Description
gateway_url	HTTPS URL of the 11Mirror gateway /mcp endpoint, for example https://connect.neuu.in/mcp
gateway_api_key	Bearer token for the gateway

OpenClaw's bundle loader only supports stdio MCP transports, so the plugin ships bin/mcp-stdio-bridge.js — a zero-dependency Node script that reads MCP JSON-RPC from stdin and relays it to the remote HTTP gateway with the bearer token. The bridge self-configures: it reads gateway_url / gateway_api_key directly from ~/.openclaw/openclaw.json under plugins.entries.11mirror.config, so no env-var interpolation is needed. Requires Node 18+ on the host.

Capabilities

Knowledge Brain (People & Process)

• Query people, relationships, contacts
• Search playbooks, SOPs, operational processes
• Retrieve meeting notes, decisions, action items
• Store new process knowledge

Structured Data (Enterprise Graph)

• Query financial records, accounting data
• Search vendor master data, supplier info
• Analyze org structure, department hierarchies
• Ingest structured business data

Skills

This plugin ships 41 skills (as of v0.26.6-fbrain.1) covering:

• Brain operations — query, search, enrichment, ingestion, citation fixing
• Content capture — ideas, articles, media, meetings, voice notes, signals
• Maintenance — dream cycle, orphan detection, frontmatter guard, testing
• Operations — task management, briefing, cron scheduling, reports, minion orchestration

See skills/RESOLVER.md for the full skill dispatch table.

Tools behind the MCP gateway

The gateway exposes 47 tools through a single /mcp endpoint:

• 44 fbrain tools for people, playbooks, meeting notes, and prose knowledge. Primary ones: query, search, get_page, list_pages, put_page, traverse_graph, get_backlinks, get_chunks. Write via put_page, add_link, add_tag, add_timeline_entry.
• 3 cognee tools for the structured-data knowledge graph: graph_query (synthesized answer via GRAPH_COMPLETION), entity_search (raw chunk hits via CHUNKS), knowledge_ingest (text → add + cognify pipeline).

Call fbrain_route first in every session — it returns the live routing protocol. Tool schemas are always discoverable via MCP tools/list.

Cognee tool naming vs upstream cognee-mcp

Upstream cognee-mcp exposes a deliberately abstracted "Minimal Memory API" (remember, recall, forget) over @mcp.tool(). 11Mirror proxies cognee's native HTTP surface (/api/v1/search, /api/v1/add, /api/v1/cognify) and names the tools after their domain role: graph_query, entity_search, knowledge_ingest. If you are reading cognee docs that reference remember/recall/forget, the mapping is:

cognee-mcp (upstream)	11Mirror gateway	Native HTTP
remember	knowledge_ingest	POST /api/v1/add then POST /api/v1/cognify
recall (graph)	graph_query	POST /api/v1/search with search_type: GRAPH_COMPLETION
recall (chunks)	entity_search	POST /api/v1/search with search_type: CHUNKS
forget	not proxied	cognee REST delete endpoints

Cognee dataset naming gotcha

The 11Mirror librarian stores content under {tenant}__{dataset}, so the cognee dataset name an agent queries is not what a human typed when running librarian-ingest. Current production datasets on the primary deployment:

• netsuite-schema-data (schema DB ingested direct, no tenant prefix)
• default__netsuite-rules (4 YAML rule files ingested via the file subcommand)

Pass these names as dataset_name on graph_query / entity_search. Calling these tools with no dataset returns an error.

Configuration

Add additional data source MCP servers to .mcp.json:

Category	Purpose
erp-accounting	ERP system for GL, subledger, JE data
data-warehouse	Financial queries and historical data
spreadsheets	Workpaper generation

Routing Protocol