nexatron-mcp
MCP server for Nexatron. Ask questions about your enterprise data from Claude and other MCP clients.
Add Nexatron from Claude's connector directory
Nexatron is listed in Claude's connector directory as a Community connector. In Claude, open the connector directory, search for Nexatron, choose Connect, and sign in to your Nexatron workspace. Nothing to install and no API key to copy. A workspace admin approves the app before members use it.
Setup guide: docs.nexatron.io/integrations/claude.
Two ways to connect
- Hosted endpoint (recommended):
https://api.nexatron.io/api/v1/mcp, Streamable HTTP with OAuth 2.1. This is what the Claude directory listing uses, and other remote-capable clients (Cursor, VS Code) can add the same URL. It has up to ten tools (seven analytics, three action), filtered by workspace entitlement and the scopes you grant. A read-only grant sees the seven analytics tools:ask,list_sources,list_metrics,explain_metric,query_metric,compare_metric_across_sources(several metrics in one call) andaudit. The action tools arelist_actions,dry_runandexecute_action. See docs.nexatron.io/quickstart/mcp. - Local stdio package (this package):
uvx nexatron-mcp, authenticated with annxa_API key. A REST-backed set of 12 read-only tools for self-hosted Nexatron instances and local use; see Available tools. The two surfaces have different tools and are versioned independently.
The rest of this README covers the local stdio package.
Install
pip install nexatron-mcp
# or
uvx nexatron-mcp
Setup
Claude Desktop: manual config
Add this to claude_desktop_config.json. On macOS it is at
~/Library/Application Support/Claude/claude_desktop_config.json; on Windows,
%APPDATA%\Claude\claude_desktop_config.json.
{
"mcpServers": {
"nexatron": {
"command": "uvx",
"args": ["nexatron-mcp"],
"env": {
"NEXATRON_API_KEY": "nxa_your_api_key",
"NEXATRON_URL": "https://api.nexatron.io/api/v1"
}
}
}
}
Restart Claude Desktop. The Nexatron tools then appear.
Claude Desktop: local extension bundle (optional)
You can also build an unsigned MCPB bundle of this local package from manifest.json. Claude asks for your API key in a masked field, so the key does not land in a config file on disk. The bundle is local-only: it is not published, and Claude's connector directory does not accept MCPB bundles. Build it with:
bash scripts/build-mcpb.sh
That writes dist/nexatron-mcp-<version>.mcpb; open it with Claude Desktop to install.
Claude Code
claude mcp add nexatron \
--env NEXATRON_API_KEY=nxa_your_key \
--env NEXATRON_URL=https://api.nexatron.io/api/v1 \
-- uvx nexatron-mcp
Options for claude go before --; everything after it is the command that starts the server.
Cursor / VS Code
Add to .cursor/mcp.json (project) or ~/.cursor/mcp.json (global), or to your VS Code MCP settings:
{
"mcpServers": {
"nexatron": {
"command": "uvx",
"args": ["nexatron-mcp"],
"env": {
"NEXATRON_API_KEY": "nxa_your_key",
"NEXATRON_URL": "https://api.nexatron.io/api/v1"
}
}
}
}
Environment Variables
| Variable | Required | Default | Description |
|---|---|---|---|
NEXATRON_API_KEY |
Yes | (none) | Your Nexatron API key (starts with nxa_) |
NEXATRON_URL |
No | https://api.nexatron.io/api/v1 |
Nexatron API URL |
Available tools (local stdio package)
These are the 12 tools the nexatron-mcp stdio package exposes against the Nexatron REST API. The hosted endpoint has a different set; see Two ways to connect.
| Tool | Description |
|---|---|
query |
Ask questions about your data in plain English |
list_connections |
See all data source connections |
list_metrics |
View certified metrics in the semantic layer |
list_views |
View logical views |
get_schema |
Explore database schemas |
test_connection |
Test data source connectivity |
detect_schema_drift |
Find schema changes |
search_knowledge_base |
Search uploaded documents (RAG) |
list_conversations |
View recent query history |
get_report |
Retrieve a saved report |
get_credit_balance |
Check remaining query credits |
whoami |
Show your user info and organization |
Example
After setup, ask Claude:
"Use Nexatron to show me total revenue by product category for Q4"
Claude calls the query tool, which sends your question through Nexatron and returns the answer with the generated query and results.
Local SSE mode (development)
NEXATRON_API_KEY=nxa_... nexatron-mcp --sse --port 8001
This starts a local SSE server on 127.0.0.1:8001 that uses your API key for every request,
so it guards itself:
- Every request needs
Authorization: Bearer <local token>. WithoutNEXATRON_MCP_LOCAL_TOKENset, a token is generated at start and printed once to your terminal (stderr); set the variable to a long random value (24+ characters) to choose your own. - A browser request from any origin other than the server's own address is refused, and the
Hostheader must be the server's address (DNS-rebinding protection). --hostother than a loopback address is refused unless you setNEXATRON_MCP_LOCAL_TOKENyourself: everyone who has that token acts as you.- A wildcard
--host(0.0.0.0,::) is refused unless you name each address clients use with--allowed-host HOST[:PORT](repeatable; the port defaults to--port). TheHosta client sends has to match one of them exactly.--allowed-hostalso adds addresses to a specific--host, such as a DNS name for it. An IPv6 address is written in brackets in client URLs, for examplehttp://[::1]:8001/sse.
It is for development on your own machine. It is not the hosted remote server; remote clients
should use https://api.nexatron.io/api/v1/mcp.
Development
Nexatron is a commercial product and its source repository is private, so there is no public clone URL. To work against the package locally, install it from source:
pip download nexatron-mcp --no-binary :all: --no-deps
tar xf nexatron-mcp-*.tar.gz && cd nexatron-mcp-*
pip install -e .
nexatron-mcp
dotSolved customers with a source agreement can contact ai@nexatron.io for
repository access.
Privacy Policy
Full policy: https://dotsolved.ai/privacy. Summary of how this connector handles data:
What is collected. The server sends the question or arguments you pass to a tool, plus
your NEXATRON_API_KEY, to the Nexatron instance at NEXATRON_URL. It collects nothing
else: no conversation history, no chat transcripts, no files, and no telemetry of its own.
It does not read Claude's memory, conversation summaries, or user files.
How it is used and stored. Requests are processed by your Nexatron instance to answer the question and are recorded in that instance's per-tenant audit trail, which exists so your own administrators can review who asked what. Query results are returned to the caller and are not retained by this package, which holds no local database and writes no data to disk; the API key is read from the environment and kept only in memory for the life of the process.
Third-party sharing. dotSolved does not sell or rent this data. Your Nexatron instance routes questions to whichever LLM provider that instance is configured to use; on a bring-your-own-LLM deployment that is your own Azure OpenAI, Bedrock, or Anthropic contract, and no question transits a dotSolved-default provider. Self-hosted deployments send nothing to dotSolved at all.
Retention. Governed by your Nexatron instance's tenant configuration, which its administrators control. This package imposes no retention of its own.
Contact. Data updates and removals: privacy@nexatron.io. General privacy questions,
including requests for a copy of your personal information: info@nexatron.io. Security
disclosures: security@nexatron.io (see SECURITY.md). These are the
addresses named in the full policy linked above.
License
MIT
Release and registry runbook
Publication is a human-authorized operation owned by nexatron-deploy. GitHub Actions is
retired, and a merge or tag in this repository never publishes anything.
1. PyPI: publish nexatron-mcp
The nexatron-mcp Python package must be on PyPI before the MCP Registry record that
references it, because the registry checks the package and the mcp-name marker in
this README.
Pre-flight checklist
- Required gates in
AGENTS.mdpass. - The version in
pyproject.tomlmatchesserver.json(both fields),manifest.json,release/nexatron-mcp-manifest.jsonand the pins incontracts/producer-gate-plan.v1.json.__version__is read from installed metadata, sopyproject.tomlis its only source. -
CHANGELOG.mdhas an entry for the version being published.
Release-candidate steps
cd server
uv build
uv run --with twine twine check dist/*
shasum -a 256 dist/*.whl dist/*.tar.gz
Record the source commit, test evidence and SHA-256 digests, then hand the candidate to the
authorized nexatron-deploy workflow. Afterwards, verify https://pypi.org/project/nexatron-mcp/.
2. Official MCP Registry
The MCP Registry is separate from Claude's connector directory. Source metadata is
server.json. The registry needs a new version to change metadata, so bump the version
with every metadata change. After the matching PyPI version is live, the owner republishes
from server/ with mcp-publisher login dns then mcp-publisher publish, signed in
with DNS authentication for nexatron.io (mcp-publisher login dns --domain nexatron.io),
which controls the io.nexatron namespace. Never publish under a personal GitHub namespace.
Do not submit a community-server PR to modelcontextprotocol/servers.
3. Third-party registries (Cursor, Smithery, Glama, mcp.so, PulseMCP)
Follow listing/SUBMISSION-RUNBOOK.md. The Cursor Directory uses a web form; its field sheet is listing/cursor-directory-submission.md.
4. Claude's connector directory
Nexatron is listed as a Community connector (published 2026-09-27). Manage the listing at https://claude.ai/directory/manage/nexatron; do not create a new submission. Evidence and remaining work are in listing/CLAUDE-DIRECTORY-READINESS.md.
5. npm: @nexatron/create-mcp (not published)
The scaffolding CLI in bootstrap/ is not published to npm, so npx @nexatron/create-mcp
does not work for users yet. Do not reference it in user-facing copy until it is published.
Pre-flight before a first publish:
-
cd bootstrap && npm testpasses. - The version in
package.jsonis correct. -
bin.create-nexatron-mcppoints to./src/index.js, whose first line is#!/usr/bin/env node. -
npm pack --dry-runshows only the intended files.
Publishing needs an npm account with rights to the @nexatron scope and is an owner action.
6. Status
| Surface | Status | Notes |
|---|---|---|
PyPI (nexatron-mcp) |
Published | 1.0.2 is the current version (1.0.0 June 2026, 1.0.1 November 2026). |
MCP Registry (registry.modelcontextprotocol.io) |
Published | io.nexatron/mcp from 1.0.2. The earlier io.github.sravanmodugula-solved/nexatron-mcp record (1.0.0, 1.0.1) is deprecated. |
| Claude's connector directory | Listed (Community) | Since 2026-09-27. Manage at https://claude.ai/directory/manage/nexatron |
modelcontextprotocol/servers (legacy list) |
N/A | No longer accepts community PRs |
npm (@nexatron/create-mcp) |
Not published | See section 5 |
7. Hosted endpoint metadata
https://api.nexatron.io/.well-known/mcp/server.json is served by nexatron-api
(app/api/well_known.py), which is authoritative for the hosted endpoint. This repository's
release/nexatron-mcp-manifest.json describes the local stdio package (its 12 tools and
stdio transport) and does not feed that route.
When the local package's tools, scopes or capabilities change:
- Update
src/nexatron_mcp/server.pyandrelease/nexatron-mcp-manifest.json. - Bump the version everywhere listed in the PyPI pre-flight checklist.
- Add a CHANGELOG entry.
Metadata
Release files for nexatron-mcp 1.0.2
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| nexatron_mcp-1.0.2.tar.gz | 41.3 kB | Without provenance | Details for nexatron_mcp-1.0.2.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| nexatron_mcp-1.0.2-py3-none-any.whl | Python 3 | none | any | Details for nexatron_mcp-1.0.2-py3-none-any.whl |
Total release size: 73.1 kB
Release files / nexatron_mcp-1.0.2.tar.gz
| Download URL | nexatron_mcp-1.0.2.tar.gz |
|---|---|
| Size | 41.3 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
af9e6e6994f2cdf890a06112cb387a3b0ee806b09f321237a35fc7a781a9ed62
|
|
BLAKE2b-256 checksum How to use checksums |
c5a47901cede5e18d358d9cc2e2106d038dfb3bfead457489bf84219d68afe89
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
uv/0.12.17 {"installer":{"name":"uv","version":"0.12.17","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":null}
|
Release files / nexatron_mcp-1.0.2-py3-none-any.whl
| Download URL | nexatron_mcp-1.0.2-py3-none-any.whl |
|---|---|
| Size | 31.9 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
6b93299c4e1fd60ed4f5f5238feda6d9e9cbaeadbd6f2901563f532e0b374580
|
|
BLAKE2b-256 checksum How to use checksums |
34af8728a3d9f6194071af40be7af6aae563c36a83ef9639a21c2a7577289250
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
uv/0.12.17 {"installer":{"name":"uv","version":"0.12.17","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":null}
|
Security for nexatron-mcp 1.0.2
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.