Investigate OpenTelemetry traces, logs, and metrics and manage personal dashboards from an AI client.
Also known as labstack/pitot
- Stars
- 124
- Tools
- —
- Downloads, 30 days
- —
- Latest version
- 2026.11.7-alpha
- Last commit
- Nov 14, 2026
- License
- Apache-2.0
What is Pitot MCP?
Pitot is an MCP server. Investigate OpenTelemetry traces, logs, and metrics and manage personal dashboards from an AI client.
Summary generated from the project's README
About Pitot MCP server
Investigating OpenTelemetry traces, logs, and metrics through an AI client is what Pitot enables for developers managing observability. This single-binary solution consolidates ingestion, Parquet storage, and DuckDB analytics into one process, eliminating the need for external storage services or complex distributed stacks. It functions as a local or remote server that accepts OTLP data via gRPC or HTTP.
Users can connect external agents using the MCP server to access typed query tools. To enable conversational investigation, you must provide an Anthropic or OpenAI API key during startup. The system provides built-in dashboards for monitoring service error rates, latency, and throughput, alongside an investigation agent that translates telemetry data into plain language. Because it is a single-node tool, it is designed for environments where data stays on your local disk rather than being sent to managed cloud providers.
What you can do with it
- 1Query telemetry with natural languageAsk an AI agent to explain specific telemetry patterns or answer questions about your system's behavior.
- 2Monitor service health via dashboardsTrack critical performance indicators like error rates and latency through integrated visual tools.
- 3Connect external AI agents via MCPIntegrate your existing AI workflows with Pitot's telemetry data using the Model Context Protocol.
Written from the project's README. Every use case is checked against a quote from it.
How to install Pitot MCP
Pitot gives each account its own endpoint, so there is no address to copy here. Get yours from the provider, put it in place of the part in braces, and add it to your client as a remote server.
https://{pitot_host}/mcp
Configuration
OTEL_EXPORTER_OTLP_ENDPOINTOTEL_EXPORTER_OTLP_HEADERSOTEL_EXPORTER_OTLP_PROTOCOLPITOT_AUTH_CODE_SECRET
| Name | Description | |
|---|---|---|
| OTEL_EXPORTER_OTLP_ENDPOINT | — | |
| OTEL_EXPORTER_OTLP_HEADERS | — | |
| OTEL_EXPORTER_OTLP_PROTOCOL | — | |
| PITOT_AUTH_CODE_SECRET | — |
Automated checks
Ran Nov 13, 2026Full report →Passed our automated checks. That is not a security audit: read the code and give the server the narrowest access it needs. How checks work
- Source repository linked and reachablePass
- Open source license (OSI)Pass
- Commit or release in the last yearPass
- Package name not close to a popular packageNot checked
- Remote endpoint uses HTTPSPass
- Destructive tools are labelledNot checked
- No hidden instructions to the modelPass
- No critical advisory in the package or its dependenciesNot checked
README
View on GitHubPitot
Pitot (pronounced PEE-toh).
Observability your agents can operate. Send OpenTelemetry to one binary, then investigate with Pitot's built-in agent or connect your own over MCP.
Current Pitot interface with deterministic offline sample telemetry.
Pitot keeps OpenTelemetry traces, logs, and metrics in one binary and one data directory, with no external storage services to operate. Pitot is pre-release and changes without migration paths; the browser client, built-in agent, MCP tools, dashboards, and alerts are preview capabilities (status). Start with the documentation or run it below.
- OTLP ingest over gRPC and HTTP, with local Parquet storage and DuckDB analytics.
- Explore and PitotQL for querying telemetry without an AI provider.
- An investigation agent that answers questions about your telemetry in plain language.
- An MCP server for external agents using the same typed query tools.
- Dashboards across traces, logs, and metrics, and alerts on service error rate, latency, throughput, and log volume.
Quick start
Docker
docker run --name pitot --restart unless-stopped -p 7520:7520 \
-v pitot-data:/var/lib/pitot/data \
-e PITOT_AUTH_CODE_SECRET=$(openssl rand -hex 32) \
labstack/pitot:latest
Open the one-time setup URL printed by the container and create the first administrator. Pitot
displays the ingest token exactly once, unless PITOT_INGEST_TOKEN already set it; save it with your collector secrets. For later sign-in
without SMTP, mint a 15-minute, single-use link against the running container's control database:
docker exec pitot pitot login-link admin@example.com
The UI, API, MCP, and both OTLP transports share port 7520. See
installation for image defaults and bind-mount ownership, and
first boot for setup and SMTP login.
Native binary
Install a checksum-verified release for Linux or macOS, amd64 or arm64:
curl -fsSL https://raw.githubusercontent.com/labstack/pitot/main/scripts/install.sh | sh
export PITOT_AUTH_CODE_SECRET=$(openssl rand -hex 32)
pitot
Set PITOT_VERSION to a release tag, such as v2026.11.0-alpha, to pin a release, and PITOT_PREFIX to choose the installation
directory. Native installs bind 127.0.0.1:7520; set PITOT_ADDR=:7520 for remote access, then
follow the same first-boot setup.
Send telemetry
A standard OTLP/HTTP exporter can use the saved ingest token:
export OTEL_EXPORTER_OTLP_PROTOCOL=http/protobuf
export OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:7520
export OTEL_EXPORTER_OTLP_HEADERS="authorization=Bearer%20$INGEST_TOKEN"
For a Collector on the same private container network, the forwarding side is:
exporters:
otlp_http/pitot:
endpoint: http://pitot:7520
headers:
Authorization: "Bearer ${env:INGEST_TOKEN}"
service:
pipelines:
traces: { receivers: [otlp], exporters: [otlp_http/pitot] }
metrics: { receivers: [otlp], exporters: [otlp_http/pitot] }
logs: { receivers: [otlp], exporters: [otlp_http/pitot] }
Use an existing otlp receiver, set INGEST_TOKEN, and replace pitot with the Collector's
reachable private hostname. OTLP/gRPC uses localhost:7520 locally; see send
telemetry for TLS and HTTP/2 setup.
Enable chat
Select PITOT_AI_PROVIDER=anthropic and inject ANTHROPIC_API_KEY, or select
PITOT_AI_PROVIDER=openai and inject OPENAI_API_KEY, then restart Pitot; the selected provider
requires its key at startup. With no provider, chat is disabled and the other features still work.
See AI settings.
Connect an MCP client
Set PITOT_PUBLIC_URL to your stable, externally reachable HTTPS origin, then point a remote MCP
client at that origin's /mcp endpoint. The client discovers and registers through OAuth; you
authorize its scopes in the browser. Pitot serves MCP 2026-07-28 and, through the initialize
handshake, 2025-11-25 and 2025-06-18 over Streamable HTTP. See connect over
MCP for setup and permissions.
How it compares
| If you use | Where Pitot differs |
|---|---|
| Grafana with Loki, Tempo, and Mimir | That stack keeps a service and a query language per signal, plus object storage underneath. Pitot keeps one process, one data directory, and one typed contract across all three signals, at the cost of the horizontal scale those components are built for. |
| SigNoz | Both are OTLP-native and self-hosted. SigNoz composes a collector, ClickHouse, and query services; Pitot compiles ingest, authoritative Parquet storage, indexed trace reads, DuckDB analytics, alerting, and the browser client into one binary. |
| Jaeger | Jaeger covers traces and expects a storage backend you run separately. Pitot ingests traces, logs, and metrics into the same store, with nothing else to deploy. |
| Prometheus with Grafana | Prometheus pulls metrics and is excellent at them. Pitot accepts pushed OTLP for all three signals and is built around investigating a specific incident rather than maintaining long-range metric series. |
| Datadog, Honeycomb, Grafana Cloud | Those are managed services: someone else runs the storage, the scaling, and the upgrades, and your telemetry leaves your network to get there. Pitot is a binary you run, on data that stays on your disk. |
| An OpenTelemetry Collector piped into ClickHouse | The same shape, assembled by hand: collector, database, dashboards, and the glue between them. Pitot is that assembly as one program, with an agent and an MCP server already wired to the same query contract. |
Pitot is a single node. It has no clustering, no replication, and no object tier; a deployment that outgrows one machine's disk and CPU has outgrown Pitot.
Architecture
One process contains ingest, storage, query, alerts, the agent, MCP, and the React client. Browser and agent clients share a typed observability contract.
Parquet holds telemetry, DuckDB query state is rebuildable, and SQLite holds transactional product state. See how storage works for durable ingest, indexed reads, and compaction while reads continue.
Configuration
Configuration is environment-only: built-in defaults followed by nonempty process environment values, validated before data files or listeners open. Settings are read once at startup; every change requires a restart.
| Environment variable | Default | Purpose |
|---|---|---|
PITOT_ADDR |
127.0.0.1:7520 |
UI, API, MCP, and both OTLP transports |
PITOT_DATA_DIR |
./data |
Parquet, query state, and control SQLite |
PITOT_AUTH_MODE |
local |
Login link or SMTP, or oidc |
PITOT_AUTH_CODE_SECRET |
— | Required in local mode and for OIDC break-glass links, 32+ characters |
PITOT_AI_PROVIDER |
— | Empty disables chat; anthropic or openai enables it |
ANTHROPIC_API_KEY |
— | Required for Anthropic chat |
OPENAI_API_KEY |
— | Required for OpenAI chat |
PITOT_RETENTION_DAYS |
30 |
Telemetry retention window |
PITOT_MCP_ENABLED |
true |
Serve the MCP endpoint at /mcp |
Every secret accepts <NAME>_FILE instead of its value; see the settings
reference for validation and runtime secret
delivery for deployment examples.
Advanced DuckDB sizing
Pitot sizes DuckDB automatically; measured workloads can override memory, connections, and threads as described in DuckDB sizing.
For TLS, health checks, backups, upgrades, retention, and recovery, see the operator runbook.
Performance
Pitot v2026.11.0-alpha accepted 507,311 mixed OpenTelemetry items/s over 306.8 seconds on a Hetzner Cloud ccx33 (8 dedicated vCPUs, 32 GB RAM), with telemetrygen on a separate ccx33. The ingest run on 6 November 2026 passed every quality gate. See the performance methodology. Pitot Bench reproduces the measurement on your hardware.
Build from source
- Go 1.27.3 and a C/C++ toolchain on native Linux or macOS, amd64 or arm64.
Build with
just buildorbash scripts/with-duckdb.sh go build ./cmd/pitot; the wrapper supplies DuckDB's pinned headers, static libraries, and CGO flags. The first build needs network access to download the checksum-verified engine. - Bun — compiles the browser assets
- Node.js 24.19.0 — runs development helpers and builds the documentation site
- just — task runner
- A 32-character authentication code secret
git clone https://github.com/labstack/pitot.git
cd pitot
just install # browser dependencies and git hooks
just build # browser assets, then the binary
Run it with the minimum configuration:
export PITOT_AUTH_CODE_SECRET=$(openssl rand -hex 32) # must be 32+ characters
./bin/pitot
Open http://localhost:7520 and follow first-boot setup. SMTP and chat are optional; without SMTP,
run ./bin/pitot login-link admin@example.com with the same configuration and data directory.
Development
Run just to list recipes, just check for the full gate, just test for Go tests, and just ui
to rebuild embedded browser assets. See CONTRIBUTING.md for environment setup,
generated assets, hooks, and project layout.
Releases
Versions are CalVer — vYYYY.M.N[-alpha|-beta|-rc], numbered from 0 within each month across all
maturity suffixes. Pitot is alpha (for example, v2026.11.0-alpha); GA has no suffix.
Container tags omit the leading v, and every release moves latest. Tagged releases publish
identical manifests to Docker Hub (labstack/pitot) and GHCR (ghcr.io/labstack/pitot); GHCR is
canonical and also carries development images.
| Image tag | Points at |
|---|---|
labstack/pitot:latest |
the newest release, mirrored on Docker Hub |
labstack/pitot:2026.11.0-alpha |
that exact release, mirrored on Docker Hub |
ghcr.io/labstack/pitot:latest |
the newest release |
ghcr.io/labstack/pitot:2026.11.0-alpha |
that exact release |
ghcr.io/labstack/pitot:main |
the tip of main |
ghcr.io/labstack/pitot:sha-<commit> |
one specific commit |
See docs/release.md for platforms, checksums, provenance, and verification.
Contributing
Issues and pull requests are welcome; follow CONTRIBUTING.md and run just check
and just test-race before opening a pull request. Report security issues through
SECURITY.md.
Scope
This repository builds Pitot without another repository; LabStack's deployment configuration,
uptime monitoring, and public demo instance are not included. The documentation site in
site/ is included, with settings pages generated from the same internal/config type the
loader binds and checked by just check.
License
Apache-2.0 © LabStack LLC. See NOTICE and THIRD_PARTY_NOTICES for attribution, and TRADEMARK for use of the Pitot name and logo.
Pitot MCP questions
How does an MCP client connect to Pitot?
Set PITOT_PUBLIC_URL to your stable HTTPS origin, then point a remote MCP client at that origin's /mcp endpoint; the client registers through OAuth and you authorize scopes in the browser.
What telemetry signals does Pitot accept?
Pitot ingests OpenTelemetry traces, logs, and metrics over OTLP via gRPC and HTTP and stores them locally.
Can I disable the MCP endpoint in Pitot?
Yes, set PITOT_MCP_ENABLED to false; it defaults to true and serves the endpoint at /mcp.
Which MCP protocol versions does Pitot support?
Pitot serves MCP 2026-07-28 plus 2025-11-25 and 2025-06-18 through the initialize handshake over Streamable HTTP.
Is Pitot a single node or a cluster?
Pitot is a single node with no clustering, replication, or object storage tier.