Adapt is a FastAPI server that turns files in a directory into APIs and UIs.
- Datasets (
.csv,.xlsx,.xls,.parquet) become API endpoints and DataTables UIs - Legacy
.xlsworkbooks are read-only. Modern.xlsxworkbooks support CRUD operations. - Markdown/HTML become browsable pages
- Media files become streaming endpoints, plus player and gallery UIs
- Python files can register custom routers
- Everything is searchable in one place via full-text
/search - Everything is reachable by agentic tools via an MCP server at
/mcp
pip install adapt-server
adapt addsuperuser --username admin /path/to/docroot
adapt serve /path/to/docroot
# Generate permissions for all discovered resources
adapt admin create-permissions /path/to/docroot __all__
# Everything below here can be done in the admin UI at
# http://localhost:8000/admin/ after logging in with the superuser account.
#
# Create a regular user
adapt admin create-user --username editor --password secret /path/to/docroot
# Reset an existing password and revoke that user's browser sessions
adapt admin change-password --username editor /path/to/docroot
# By default, the editor user has no permissions.
# See available groups (created by `adapt admin create-permissions`) and assign user to desired group
adapt admin list-groups /path/to/docroot
adapt admin add-to-group --username editor --group <group_name> /path/to/docrootUseful URLs:
/landing page/admin/admin UI/api/<resource>resource API/ui/<resource>resource UI/schema/<resource>resource schema/searchfull-text search across every resource you can read/mcpMCP server for agentic tools (see MCP Interface below)
From files in your docroot, Adapt auto-discovers resources and mounts routes with extensionless URLs where possible.
Example:
data/
employees.csv
sales.xlsx
video.mp4
readme.md
stats.py
Rough output:
/api/employees,/ui/employees,/schema/employees/api/sales/<sheet>,/ui/sales/<sheet>/media/video.mp4,/ui/video.mp4,/ui/media/readme/api/stats/*
This reflects the current implementation in the codebase.
- Authentication: session cookies, API keys (
X-API-Key), and inactive-user enforcement - Authorization: RBAC (users, groups, permissions), plus superuser bypass
- Password security: PBKDF2 hashing with per-user salts
- Password changes: self-service and administrator resets revoke all browser sessions for the user
- Session security: expiration enforcement, sliding renewal, cleanup task
- CSRF protection: enforced for cookie-authenticated unsafe methods (
POST/PUT/PATCH/DELETE), and covers mixed session and API-key requests - Redirect hardening: login
nextpaths are validated as local relative paths - Response hardening: CSP,
X-Frame-Options,X-Content-Type-Options,Referrer-Policy,Permissions-Policy, HSTS (when TLS is enabled) - Host header hardening: Trusted Host middleware
- Data integrity: lock-based, atomic writes for mutable dataset plugins
- Auditability: audit records for authentication, administration, and successful dataset mutations
- Sensitive response cleanup: admin user APIs no longer expose
password_hash
- Use TLS in non-local environments (
--tls-certand--tls-key). TLS makes secure cookies and HSTS protections effective. - API-key-only clients are exempt from CSRF checks by design. Browser flows that use cookie authentication need CSRF tokens.
- Adaptive discovery and route generation
- Dataset CRUD with schema exposure
- Caching with invalidation on mutations
- Built-in admin UI for users/groups/permissions/locks/cache/api keys/audit logs
- Root-level file upload endpoint (
POST /api/uploads) with permission checks and audit logging - Optional landing-page upload card for authenticated users with root write permission
- Plugin architecture with companion overrides in
.adapt/ - Permission-filtered full-text search across every resource type
- MCP server for agentic tool access, mounted alongside the REST API
Upload permission note:
- Non-superusers need
writepermission on the document-root boundary. - In admin permission creation, use an empty resource (or
__root__) with actionwrite. Then assign that permission through a group.
GET /search?q=<query> searches datasets, Markdown, HTML, and media metadata
in one ranked list. The search filters results to what the caller can read.
A query term that matches a resource you cannot see does not appear in the
results. It also does not appear in the result count.
curl -H "X-API-Key: <key>" "http://localhost:8000/search?q=parental+leave"The index refreshes step by step on startup (search_on_startup, default
true). You can also rebuild it on demand with adapt reindex <root>. See
the API Reference for query
parameters and the result shape.
Adapt mounts a Model Context Protocol
server at /mcp. This server runs on the same host and port as the rest of
Adapt. It exposes five tools: list_resources, get_schema,
read_resource, write_resource, and search. Each tool uses the same
permission checks and plugin methods as the REST API. There is no separate
process, no separate API surface, and no extra permission model to
maintain.
Minimal walkthrough — create an account for the agent, grant it read access, mint an API key, and connect a client:
adapt addsuperuser /path/to/docroot --username admin
adapt serve /path/to/docroot &
adapt admin create-permissions /path/to/docroot __all__
adapt admin create-user /path/to/docroot --username agent --password <strong-password>
adapt admin add-to-group /path/to/docroot --username agent --group <resource>_readonlyLog in as agent. From /profile, issue an API key for yourself. Any
authenticated user can create an API key for themself. A superuser is not
necessary for this step. Then point a client at /mcp with that key:
# Claude Code CLI
claude mcp add --transport http adapt http://localhost:8000/mcp \
--header "X-API-Key: <key>"// Generic MCP client config (Claude Desktop and similar)
{
"mcpServers": {
"adapt": {
"url": "http://localhost:8000/mcp",
"headers": { "X-API-Key": "<key>" }
}
}
}MCP does an authentication check when a tool runs. Tool calls use the shared
authentication resolver (session cookie, API key, or Bearer JWT when OIDC is
on). API keys remain the simple option for scripts. OAuth MCP clients use
Bearer tokens. Set mcp_enabled: false in .adapt/conf.json (or
ADAPT_MCP_ENABLED=false) to remove /mcp entirely. For setup and
troubleshooting, read the MCP guide.
For dataset reads, sort is the column name and order must be asc or
desc.
For dataset endpoints, write operations use this payload structure:
{
"action": "create|update|delete",
"data": []
}For update and delete, include the fields the operation needs, for
example _row_id.
adapt serve <root> [--host ... --port ... --tls-cert ... --tls-key ... --reload --readonly --debug]
adapt check <root>
adapt addsuperuser <root> --username <name>
adapt list-endpoints <root>
adapt reindex <root> [--force]
adapt admin list-resources <root>
adapt admin create-permissions <root> __all__Use --reload during development. Uvicorn watches Python files in the document
root and restarts Adapt after a change.
A container image is published to ghcr.io/mcindi/adapt-server, and a Helm
chart is published to oci://ghcr.io/mcindi/charts/adapt:
docker run -p 8000:8000 -v "$(pwd)/docroot:/data" ghcr.io/mcindi/adapt-server:latest
helm install myadapt oci://ghcr.io/mcindi/charts/adapt --version 0.5.2The Helm chart README and the Kubernetes manual cover persistence, uploads, admin bootstrap, and service exposure.
Read the full documentation at https://www.mcindi.com/adapt/.
More documentation is under docs/manual/.
- Manual index: docs/manual/index.md
- Security: docs/manual/security.md
- Quick start: docs/manual/quick_start.md
- Configuration: docs/manual/configuration.md
- API reference: docs/manual/api_reference.md
- MCP guide: docs/manual/mcp_guide.md
- Plugin development: docs/manual/plugin_development.md
- Known limitations: docs/manual/known_limitations.md
- Deployment (container and Kubernetes): docs/manual/deployment.md
- Release guide: RELEASING.md
Generated reference documentation is under docs/reference/. MkDocs builds
this documentation, and GitHub Pages publishes it.
- REST API reference: generated from app routes with an empty docroot
- OpenAPI schema artifact: generated from that same common-surface schema
- Python API reference: generated from docstrings and signatures
Build the documentation locally:
python -m pip install -e ".[dev]"
python -m pip install -r requirements-docs.txt
mkdocs build --strictMIT. See LICENSE.