Pinot MCP Server

Table of Contents

• Overview

• Features

• Quick Start

• Configuration Reference

• Docker Build

• Claude Desktop Integration

• Try a Prompt

• Security and Vulnerability Reporting

• Developer Notes

• Repo Maintainer

Related MCP server: mcp-oceanbase

Overview

This project is a Python-based Model Context Protocol (MCP) server for interacting with Apache Pinot. It is built using the FastMCP framework. It is designed to integrate with Claude Desktop to enable real-time analytics and metadata queries on a Pinot cluster.

It allows you to

• List tables, segments, and schema info from Pinot

• Execute read-only SQL queries

• View index/column-level metadata

• Designed to assist business users via Claude integration

• and much more.

Features

• Every tool advertises typed input and output JSON Schemas, MCP risk annotations, and failure-recovery guidance for agent planning.

• Large query, table, segment-name, and segment-metadata responses use bounded pages with continuation metadata instead of returning unbounded agent context.

• Read-only SQL is parsed and enforced before execution; validation, permission, and transient connectivity errors are surfaced as actionable MCP errors.

• Every mutating tool supports dry_run; always preview the exact target and payload before applying. Applying requires the preview's short-lived, one-time confirmation_token, including for table-filter reloads. A preview is not a guarantee that Pinot will accept the later write.

• Single-purpose schema and table-config inspection tools avoid ambiguous combined operations: use get_schema and get_table_config independently.

MCP Tool Contract

Tool names are case-sensitive and use underscores. Version 4 renamed four tools to make every operation verb-first; clients using the former noun-first names must update their calls.

For every schema, table-config, or table-filter change, first call the same tool with dry_run=true, present the preview to the user, and call it with dry_run=false and the preview's one-time confirmation_token only after confirmation. Editing a table-filter file after preview invalidates its token. Pinot performs authoritative validation during table/schema apply calls, so a write can still fail after a successful preview.

Query execution evidence

read_query returns the existing page fields plus metadata. For example, {"query":"SELECT service, count(*) FROM events GROUP BY service","limit":20} returns metadata.completeness as complete, partial, or unknown, along with native server counts, query ID, query SHA-256, and bounded execution statistics. Missing resultTable or required execution flags remains unknown. A partial server response or group limit remains partial; an empty page is not proof of healthy telemetry. Native complete describes this bounded SQL execution, not coverage of the incident window or the entire dataset. row_limit_reached is separate from execution completeness and paging.

The MCP query path uses pinotdb>=9.2.1 through cursor.execute_native() for submission and decoding. The SDK exposes full structured query_statistics; MCP projects bounded counters and safe execution evidence into the tool response. The path submits once and does not switch transports after an ambiguous failure. Its unique clientQueryId and applicationName=mcp-pinot support optional correlation with query history or broker logs when available. Python integrations can call execute_query_with_metadata(..., timeout_seconds=5) to bound native and HTTP timeouts without modifying shared configuration. HTTP timeouts are inactivity limits; late results are rejected, but this is not native cancellation.

Tool results, including tool errors, include timing evidence in the standard MCP response _meta["io.github.startreedata/mcp-pinot"]. Existing tool data and output schemas are preserved. For example, this response metadata fragment requires no log access:

{
  "_meta": {
    "io.github.startreedata/mcp-pinot": {
      "request_id": "38cb1444b20c446a96bc272531f34518",
      "status": "success",
      "admitted": true,
      "queue_wait_ms": 12,
      "execution_ms": 84,
      "duration_ms": 97
    }
  }
}

MCP rate-limit protocol errors preserve their JSON-RPC code and message and carry the same envelope in error.data._meta["io.github.startreedata/mcp-pinot"]. Other transport/protocol failures or cancellation may not produce a tool result.

With MCP_TRANSPORT=http on the default port 8080, a FastMCP Python client reads wire _meta through result.meta. call_tool_mcp also returns tool errors:

from fastmcp import Client


async def read_timing():
    async with Client("http://localhost:8080/mcp") as client:
        result = await client.call_tool_mcp("read_query", {"query": "SELECT 1"})
        return result.meta["io.github.startreedata/mcp-pinot"]

Orchestrators should read .meta and forward this evidence to the agent when the host does not automatically surface protocol metadata.

queue_wait_ms measures only waiting for the MCP concurrency semaphore. execution_ms measures the admitted MCP tool pipeline, including SDK HTTP work and validation; admitted=false means that pipeline never started. duration_ms covers the server's tool invocation, including admission checks, and excludes model reasoning and client delivery. status is success or error for the MCP outcome; for read_query, separately inspect metadata.completeness.

An agent can use queue_wait_ms to identify MCP admission delay, then inspect read_query's metadata.native_stats.timeUsedMs for Pinot's native query duration when present. Native query time overlaps execution_ms; their different timer boundaries do not support exact network-overhead subtraction. The shared request_id also appears in query metadata and native clientQueryId, enabling optional history/log correlation when more context is needed.

Response timing metadata contains no secrets, principal identifiers, tool arguments, SQL, or rows. Audit logs record the same timings with hashed principal identifiers. MCP_MAX_RESPONSE_BYTES limits the tool payload before this bounded timing envelope is added; it excludes the added observability metadata. Cancellation is audit-only when no response is returned and does not prove that Pinot execution stopped.

Incident tools are disabled until the server starts with --incident-profiles. See the incident evidence guide and example profile for configuration and a workflow using begin_investigation, query_incident, optional get_trace, and finish_investigation. Evidence queries expose the same response _meta timings and native request IDs when reported; log access is optional. Runs are owner-bound and budgeted; complete execution still leaves root-cause hypotheses and dataset coverage unvalidated. This first version uses bounded in-memory state in one process.

Pinot MCP in Action

See Pinot MCP in action below:

Fetching Metadata

Fetching Data, followed by analysis

Prompt: Can you do a histogram plot on the GitHub events against time

Sample Prompts

Once Claude is running, click the hammer 🛠️ icon and try these prompts:

• Can you help me analyse my data in Pinot? Use the Pinot tool and look at the list of tables to begin with.

• Can you do a histogram plot on the GitHub events against time

Quick Start

Prerequisites

Install uv (if not already installed)

uv is a fast Python package installer and resolver, written in Rust. It's designed to be a drop-in replacement for pip with significantly better performance.

curl -LsSf https://astral.sh/uv/install.sh | sh

# Reload your bashrc/zshrc to take effect. Alternatively, restart your terminal
# source ~/.bashrc

Installation

# Clone the repository
git clone https://github.com/startreedata/mcp-pinot.git
cd mcp-pinot
uv pip install -e . # Install dependencies

# For development dependencies (including testing tools), use:
# uv pip install -e .[dev] 

Configure Pinot Cluster

The MCP server expects a uvicorn config style .env file in the root directory to configure the Pinot cluster connection. This repo includes a sample .env.example file that assumes a pinot quickstart setup.

mv .env.example .env

Configuration Reference

The server loads configuration from environment variables and from a .env file found from the current working directory. Process environment variables take precedence over .env, so deployment-time settings cannot be silently replaced by a checked-out file.

Common Profiles

Pinot Connection

MCP Server

Authentication

An auth provider is required before binding HTTP or HTTPS to a non-loopback host.

Table Filtering

See SECURITY.md for the production exposure checklist and vulnerability reporting process.

Configure Table Filtering (Optional)

⚠️ Security Note: For production access control, use Pinot's native table-level ACLs (available since Pinot 0.8.0+). Table filtering in this MCP server is a convenience feature for organizing tables and improving UX, not a security boundary. It uses best-effort SQL parsing and should not be relied upon for security.

Table filtering allows you to control which Pinot tables are visible through the MCP server. This is useful for:

• Reduce Cognitive Load: Focus on relevant tables when your Pinot cluster has hundreds or thousands of tables

• Multi-Tenancy UX: Run multiple MCP server instances against the same Pinot cluster, each showing different table subsets for different teams or use cases

• Environment Separation: Deploy different MCP server instances (dev, staging, prod) that show only environment-specific tables

• Hide System Tables: Filter out internal, test, or deprecated tables from end-user view

When table filtering is enabled, all table operations are filtered to show only the configured tables.

What Gets Filtered

Table filtering applies across all MCP operations:

1. Table Listing - Only configured tables appear in table lists

2. Query Execution - SQL queries are checked to ensure all referenced tables (in FROM, JOIN, subqueries, CTEs, etc.) match the configured patterns

3. Table Operations - Direct table access operations filter by table name:

   • Get table details, size, and metadata

   • Get table segments and segment metadata

   • Get index/column details

   • Get/update table configurations

4. Schema Operations - Schema operations filter by schema name:

   • Get/create/update schemas

   • Create table configurations

Setup

Copy the example configuration file:

cp table_filters.yaml.example table_filters.yaml

Edit table_filters.yaml to specify which tables to include:

included_tables:
  - production_*        # All tables starting with "production_"
  - analytics_events    # Specific table name
  - metrics_*          # All tables starting with "metrics_"

Configure the filter file path in your .env:

PINOT_TABLE_FILTER_FILE=table_filters.yaml

Pattern Matching

The filter supports glob-style patterns using standard Unix filename pattern matching:

• exact_table_name - Matches exactly this table

• prefix_* - Matches all tables starting with "prefix_"

• *_suffix - Matches all tables ending with "_suffix"

• *pattern* - Matches all tables containing "pattern"

• sharded_table_? - Matches tables with exactly one character after the underscore (e.g., sharded_table_1, sharded_table_a)

Query Filtering

When filtering is enabled, SQL queries are checked before execution:

• Supported SQL Features: FROM clauses, JOIN clauses (INNER, LEFT, RIGHT, OUTER, CROSS), subqueries, CTEs (WITH), and comma-separated table lists

• Quoted Identifiers: Supports both double-quoted ("table name") and backtick-quoted (`table_name`) table names

• Schema Prefixes: Handles schema-qualified table names (e.g., database.schema.table)

• Comments: Removes SQL comments before checking

Example filtered query:

SELECT * FROM allowed_table
JOIN other_table ON allowed_table.id = other_table.id

Error: Query references unauthorized tables: other_table. Allowed tables: allowed_table, prod_*

Configuration Features

Fail-Fast Validation:

• ⚠️ If PINOT_TABLE_FILTER_FILE is configured but the file doesn't exist, the server will fail to start with a FileNotFoundError

• This prevents accidentally showing all tables due to misconfiguration

• Empty filter files or missing included_tables key will show all tables (no filtering)

Comprehensive Filtering:

• All MCP tools that access tables apply filtering before execution

• Consistent filtering across all table access points

• Clear error messages indicate which tables don't match the configured patterns

Disabling Table Filtering

To disable table filtering, either:

1. Remove the PINOT_TABLE_FILTER_FILE environment variable, or

2. Don't configure it in your .env file

When not configured, all tables in the Pinot cluster are visible.

When a filter file supplies both allow_all: true and a non-empty included_tables, the explicit allow-list takes precedence and the server logs a warning. Applying a reload requires the token from an unchanged dry-run candidate.

Read-only Query Enforcement

The read_query tool always validates SQL before forwarding it to Pinot. It accepts one statement only, and that statement must be a read-only SELECT or WITH ... SELECT query. SQL comments are stripped, semicolon-stacked statements are rejected, and write/DDL/admin keywords are blocked.

Configure OAuth Authentication (Optional)

To enable OAuth authentication, set the following environment variables in your .env file:

Required variables (when OAUTH_ENABLED=true):

• OAUTH_CLIENT_ID: OAuth client ID

• OAUTH_CLIENT_SECRET: OAuth client secret

• OAUTH_BASE_URL: Your MCP server base URL

• OAUTH_AUTHORIZATION_ENDPOINT: OAuth authorization endpoint URL

• OAUTH_TOKEN_ENDPOINT: OAuth token endpoint URL

• OAUTH_JWKS_URI: JSON Web Key Set URI for token verification

• OAUTH_ISSUER: Token issuer identifier

Optional variables:

• OAUTH_AUDIENCE: audience tokens are validated against. Defaults to the canonical MCP resource URI (OAUTH_BASE_URL + MCP_PATH). Set it when your provider issues a different aud — for example an IdP that puts the client ID there.

• OAUTH_GRANTED_SCOPES: Pinot scopes granted to authenticated principals (default all three). Use pinot:read to make the deployment read-only for every OIDC caller.

• OAUTH_REQUIRED_SCOPES: baseline scopes an access token must already carry (default: none enforced).

• OAUTH_EXTRA_AUTH_PARAMS: Additional authorization parameters as JSON object (e.g., {"scope": "openid profile"})

Tool-level authorization uses pinot:read / pinot:write / pinot:admin. General-purpose OIDC providers issue a fixed scope catalog and cannot mint resource scopes like these, so OAUTH_GRANTED_SCOPES is what makes an authenticated user able to call anything — narrow it rather than leaving tools ungated.

Example configuration:

OAUTH_ENABLED=true
OAUTH_CLIENT_ID=client-id
OAUTH_CLIENT_SECRET=client-secret
OAUTH_BASE_URL=http://localhost:8000
OAUTH_AUTHORIZATION_ENDPOINT=https://example.com/oauth/authorize
OAUTH_TOKEN_ENDPOINT=https://example.com/oauth/token
OAUTH_JWKS_URI=https://example.com/.well-known/jwks.json
OAUTH_ISSUER=https://example.com
OAUTH_AUDIENCE=http://localhost:8000/mcp
OAUTH_EXTRA_AUTH_PARAMS={"scope": "openid profile"}

Run the server

uv --directory . run mcp_pinot/server.py

You should see logs indicating that the server is running.

Security notes:

• STDIO is the default. When HTTP is selected it binds to 127.0.0.1; set MCP_HOST=0.0.0.0 only with OAuth or static-token authentication plus TLS or an authenticated reverse proxy.

• The server refuses to start when HTTP is bound to a non-loopback host without an auth provider (AUTH_PROVIDER=oauth or static, or the legacy OAUTH_ENABLED=true).

• read_query enforces a single read-only SQL statement before execution. This is a guardrail, not a replacement for Pinot authentication and authorization.

• The supported mcp[cli] dependency includes DNS rebinding protections for the Streamable HTTP server.

• Confirmation replay state and rate-limit buckets are process-local. Run exactly one server process/Helm replica. The chart rejects replicas != 1; horizontal scaling requires a shared state-store implementation.

• /readyz reports MCP process readiness, not Pinot cluster health. Use test_connection to diagnose Pinot dependencies.

Launch Pinot Quickstart (Optional)

Start Pinot QuickStart using docker:

docker run --name pinot-quickstart -p 2123:2123 -p 9000:9000 -p 8000:8000 -d apachepinot/pinot:1.5.1 QuickStart -type batch

Query MCP Server

uv --directory . run examples/example_client.py

This quickstart just checks all the tools and queries the airlineStats table.

Claude Desktop Integration

Open Claude's config file

vi ~/Library/Application\ Support/Claude/claude_desktop_config.json

Add an MCP server entry

{
  "mcpServers": {
      "pinot_mcp": {
          "command": "/path/to/uv",
          "args": [
              "--directory",
              "/path/to/mcp-pinot-repo",
              "run",
              "mcp_pinot/server.py"
          ],
          "env": {
            // You can also include your .env config here
          }
      }
  }
}

Replace /path/to/uv with the absolute path to the uv command, you can run which uv to figure it out.

Replace /path/to/mcp-pinot with the absolute path to the folder where you cloned this repo.

Note: you must use stdio transport when running your server to use with Claude desktop.

You could also configure environment variables here instead of the .env file, in case you want to connect to multiple pinot clusters as MCP servers.

Restart Claude Desktop

Claude will now auto-launch the MCP server on startup and recognize the new Pinot-based tools.

Using the MCP Bundle

The release workflow publishes a Claude Desktop MCP Bundle (.mcpb). Its UV runtime installs the locked dependencies for the user's platform, so one small bundle works across macOS, Linux, and Windows. To build one locally:

npm install -g @anthropic-ai/mcpb@2.1.2
mcpb validate manifest.json
mcpb pack

Open the resulting .mcpb file to install it in Claude Desktop.

Security and Vulnerability Reporting

See SECURITY.md for vulnerability reporting instructions, security categories, and the checklist for safely exposing the MCP HTTP endpoint.

Developer

• MCP tool definitions live in mcp_pinot/server.py; Pinot HTTP/DB operations live in mcp_pinot/pinot_client.py.

• See MCP SDK compatibility and migration for the dependency policy, SDK 2 integration changes, validation, and rollback.

Build

Build the project with

uv sync --frozen

Test

Test the repo with:

uv run pytest --cov=mcp_pinot

Build the Docker image

docker build -t mcp-pinot .

Run the container

docker run --rm -i -v "$(pwd)/.env:/app/config/.env:ro" mcp-pinot

This uses the default STDIO transport. For HTTP/Kubernetes deployments, configure an inbound auth provider before binding to a non-loopback address; see the configuration and Helm sections above.