Concierge Core Kit
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Concierge Core Kitstart a conversation for my tenant's support agent"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Concierge Core Kit
A FastAPI backend for conversational assistants, with durable conversations, tenant-aware resources and an MCP interface.
The source includes visitor turns and streaming, agent configuration, widget versions, connector policies, audit events and PostgreSQL migrations. It is a standalone extraction of Concierge Core, not the runtime for every agent in a company and not a replacement for a workflow engine or Tower.
Start with the offline suite
Requires Python 3.12–3.14. From a source checkout:
python3 -m venv .venv
.venv/bin/python -m pip install -e '.[dev]'
.venv/bin/python scripts/test_offline.pyThe runner strips inherited credentials, ignores developer .env files and blocks real Python network connections. Tests use synthetic identities and mocked providers. The release was checked on Python 3.14: 2,174 passed; 10 database-dependent tests skipped. These numbers describe this exported suite, not a production deployment or an independent security audit.
Related MCP server: Business Assistant MCP
Architecture
flowchart LR
C["Console / visitor client"] --> H["HTTP + streaming"]
A["Authorized MCP client"] --> M["MCP tools"]
H --> P["Identity + tenant / action checks"]
M --> P
P --> D["PostgreSQL: agents, conversations, events, versions"]
P --> K["Configured KB / connectors"]
H --> R["Configured model provider"]Included | Requires your deployment |
FastAPI application and MCP tool implementations | TLS, service configuration and tenant provisioning |
PostgreSQL schema and migrations through | PostgreSQL and backup operations |
Visitor token, origin and principal checks | Identity provider, keys and correctly scoped grants |
KB and model client integrations | Compatible KB service and provider credentials |
Voice, widget and connector integration code | Their external services and client applications |
Local database verification
Use a dedicated disposable PostgreSQL 17 database. CONCIERGE_TEST_DSN is the only opt-in for tests that write to a real database; DATABASE_URL alone does not opt in.
export DATABASE_URL='postgresql://your_test_user@127.0.0.1:5432/concierge_test'
export KB_BASE_URL='https://kb.example.test'
export KB_TOKEN='synthetic-test-token'
export OPENROUTER_API_KEY='synthetic-test-key'
export ENV=test
# Emit the migration SQL, then apply it using psql.
# This path supports migrations containing multiple SQL statements.
.venv/bin/alembic upgrade head --sql > migration.sql
psql "$DATABASE_URL" -v ON_ERROR_STOP=1 -f migration.sql
export CONCIERGE_TEST_DSN="$DATABASE_URL"
.venv/bin/pytest -q tests/test_bus.py tests/test_stream_routes.py \
tests/test_widget_proof_nonce_db.py tests/test_erasure_matrix.pyRelease verification ran those 11 checks on a fresh PostgreSQL 17 database with fsync, full_page_writes and synchronous_commit enabled. They cover concurrent event sequences, notifications, streams, proof nonce use and erasure. One check overlaps the offline suite. This is not a crash-survival or exhaustive concurrency claim.
Configure a running service
Copy .env.example to .env, fill the required settings and review every authentication setting for your environment. The example contains no usable credentials; the test values above cannot operate a real assistant. Migrate the chosen database separately, then:
.venv/bin/uvicorn app.main:create_app --factory --host 127.0.0.1 --port 8000Provision tenants, agents, KB documents and caller permissions before serving conversations. /health is liveness; /ready also checks configured dependencies. The packaged persona and claims policies are synthetic examples, not company knowledge. The optional scripts/bootstrap_meta_concierge.py upserts an example tenant and agent in the configured database; inspect its model, caps and references before using it.
See the generated HTTP reference and MCP tool catalog. A listed tool is not a grant. Wiring a service token into a caller path does not establish caller authorization: review tenant, action and credential binding when integrating a KB or identity system.
Release scope
This source release excludes internal deployment configuration, customer knowledge, private differential-test references and company history. Synthetic examples replace customer fixtures and internal claims registries. MCP is constrained to the compatible 1.x SDK; Pydantic AI is pinned to the tested version. The console, KB server, Tower and production operations are separate projects.
No live model, voice, commerce, identity-provider or end-to-end client integration was exercised for this release. Do not expose an unconfigured checkout directly to the internet.
This server cannot be deployed
Maintenance
Related MCP Connectors
Hosted MCP memory and agent control plane for durable conversations, jobs, and operations.
Agent communication platform for agent to agent messaging via MCP. Messages, channels, skills.
Remote MCP for A2A caller identity, scope policy, verdict receipts, and audit history.
Let AI agents query data and act across all your business apps via MCP.
Related MCP Servers
- AlicenseBqualityNot gradedmaintenanceEnables AI assistants to interact with and manage Google Cloud Platform resources including Artifact Registry, BigQuery, Cloud Build, Compute Engine, Cloud Run, Cloud Storage, and monitoring services through a standardized MCP interface.1MIT
- FlicenseNot gradedqualityCmaintenanceMulti-tenant, stateless MCP server for conversational business assistants, handling business resolution, knowledge search, and data mutations with human confirmation and security guardrails.-
- AlicenseNot gradedqualityCmaintenanceEnables AI-driven customer support operations including conversation management, knowledge base, contacts, metrics, and settings via MCP.MIT
- AlicenseNot gradedqualityBmaintenanceConnects MCP clients to Gemini Cloud Assist APIs, enabling users to understand, manage, and troubleshoot their Google Cloud environment using natural language.67Apache 2.0