Last verified: September 14, 2026, against clawmetry 0.12.877 and the Pro extension 0.7.28
This page lists what ClawMetry connects to today and what it does not. It is a table rather than a wall of logos, because a name alone says little: a provider being listed does not mean ClawMetry collects from it, and an export does not mean ClawMetry can stop an application it receives telemetry from.
How to read it: each row says what kind of integration it is, which way data flows, what data is available, what it needs, where it falls short, and links to its documentation. Work that is finished but not usable in a released package is marked Pending release, and work that is not finished is marked Planned. Neither links setup docs. Listing a product here does not mean a partnership, an endorsement or a certification; product names belong to their owners.
Four kinds of integration are kept apart: runtime collection (ClawMetry reads an agent's own session data on its machine), telemetry ingestion (an application sends telemetry to ClawMetry), provider or billing route (how a model provider's usage is priced) and export or enforcement destination (ClawMetry sends its records elsewhere, or a pipeline acts on them).
| Integration | Kind | Direction | Data available | Requires | Limits | Status | Docs |
|---|---|---|---|---|---|---|---|
| Coding agent runtimes (Claude Code, Codex, Cursor and the rest of the supported list) | Runtime collection | The runtime's local session data, read by ClawMetry on the same machine | Sessions, tool calls and tokens. Cost where the runtime records it or a price can be derived. | ClawMetry installed where the agent runs. Which runtimes need a paid plan is on /pricing. | Reads only what each runtime writes locally, so the fields differ by runtime. | Available | What each runtime exposes |
| GitHub Copilot CLI | Runtime collection | Copilot CLI's local event log and usage ledger, read by ClawMetry | Conversations, tool calls, model routing, token split and the AI-credit cost the ledger records. | Starter plan or above. | This row is the terminal CLI. Copilot Chat in VS Code is the next row; Copilot in JetBrains and in the browser is not read. | Available | GitHub Copilot |
| GitHub Copilot Chat in VS Code, and GitHub organization Copilot usage import | Runtime collection | VS Code's own chat session files, read by ClawMetry on the same machine. The organization import is a command an operator runs against GitHub's usage reports. | Conversations answered by Copilot Chat in VS Code or VS Code Insiders: messages, tool calls with their arguments and results, model, tokens and Copilot credits. The import stores GitHub's organization and per-user daily metrics reports and AI-credit billing usage reports. | Starter plan or above and the Pro extension 0.7.28 or later. The import needs a GitHub token you supply with organization metrics and billing read access. | Pause, Stop and Kill are refused on these sessions, because a chat shares the editor's process (clawmetry 0.12.877). Cost comes from the credits VS Code records and is otherwise unknown, never estimated. Copilot in JetBrains and in the browser is not read. Imported organization figures are consumption, not joined to sessions and not an invoice. | Limited | GitHub Copilot: VS Code and organization usage |
| OpenTelemetry ingest (OTLP over HTTP) | Telemetry ingestion | Your application or runtime sends to ClawMetry at /v1/traces, /v1/metrics and /v1/logs |
The spans, metrics and logs the sender emits. Tokens and cost only when the sender includes them. | The open-source package. Requests are authenticated; an unauthenticated mode for a local-only receiver is opt-in. Binary protobuf needs clawmetry[otel]. |
In clawmetry 0.12.877, received trace spans are stored for the tracing view but are not checked by Guard detectors or redacted before storage. That change is merged for the next release, which is not published yet. Telemetry arrives after an action has run, so it cannot hold or block that action. | Limited | OpenTelemetry ingest |
| Python applications that call model APIs | Telemetry ingestion | Captured inside your Python process. In the released package the captured calls do not reach ClawMetry's store. | None in the dashboard. The released interceptor records each model call to a local JSONL file only. | Not available today as a dashboard integration. | Not in a released package. The released interceptor writes to ~/.clawmetry/intercepted.jsonl, a file the released daemon does not ingest, so captured calls do not appear in the dashboard. Capture covers the Anthropic, OpenAI, Google and OpenRouter API hosts only, from Python code using httpx or requests. |
Planned | No docs linked until ingestion ships |
| Tested recipes for LangGraph, Azure AI Foundry and AWS AgentCore | Telemetry ingestion | Not available today | None beyond the OpenTelemetry ingest row. | Not available today. | Not in a released package. No tested recipe is published for these frameworks. A LangGraph recipe is in review; Azure AI Foundry and AWS AgentCore have none in progress. | Planned | No docs yet |
| LiteLLM and other LLM gateways | Telemetry ingestion | Not available today | None in a released package. | Not available today. | Not in a released package. No gateway integration is published, so spend routed through a gateway is not attributed to teams. A LiteLLM integration is in review. | Planned | No docs yet |
| Amazon Bedrock and Azure OpenAI | Provider or billing route | No connection to the provider. Cost is derived from the model id a runtime records. | An estimated cost from ClawMetry's pricing table when the recorded model id is recognised, including Bedrock-style ids such as us.anthropic.…. |
A runtime that records the model id. | List-price estimates, not your contract or invoice rates. ClawMetry does not read these services' usage or bills, and the Python capture above does not cover their endpoints. | Limited | How cost is computed |
| Anthropic usage and cost reports valued at your contract rates | Provider or billing route | An operator command reads Anthropic's organization usage and cost reports through its Admin API | Daily usage by model, service tier and workspace, and the cost report, valued at a rate card you supply. | The Pro extension 0.7.28 or later, an Anthropic Admin API key, and a clawmetry release that includes the price book. | Not usable in a released package yet: the importer and valuation are in the Pro extension 0.7.28, but valuing usage at contract rates needs the price book, which is merged into clawmetry for the next release and not published yet. Importing OpenAI usage is not built. A usage report is not an invoice, and ClawMetry does not read provider invoices. | Pending release | No setup docs until the price book is released |
| OpenTelemetry export to your collector | Export or enforcement destination | ClawMetry pushes to your OTLP over HTTP endpoint | Agent activity, tokens and cost, next to the rest of your infrastructure telemetry. | Pro plan. CLAWMETRY_OTEL_EXPORT_ENDPOINT, with optional headers for your collector's authentication. |
Pushed on a set interval, not streamed per event. | Available | Exporting out |
| Webhooks, Slack and Discord | Export or enforcement destination | ClawMetry posts to your endpoint when an alert rule fires | The alert record as JSON. Slack and Discord receive it as a formatted message. | Pro plan: custom alert rules, custom webhooks and alert webhooks all need it. The built-in agent-attention alert (in the dashboard and on Telegram) is included on every plan. | Notification only. A webhook does not pause or stop an agent. | Available | Alert delivery |
| PagerDuty and Opsgenie | Export or enforcement destination | ClawMetry sends to the PagerDuty Events API v2 or the Opsgenie alert API | An alert with its severity mapped to the service's levels and a deduplication key. | Pro plan, and a PagerDuty routing key or an Opsgenie API key. | Opens alerts. Notification only; it does not act on the agent. | Available | PagerDuty and Opsgenie sinks |
| SIEM over syslog, for receivers such as Splunk, QRadar or Elastic | Export or enforcement destination | ClawMetry sends to your syslog collector | ClawMetry events as CEF or JSON, framed as RFC 5424 syslog over UDP, TCP or TLS. | Enterprise plan. CLAWMETRY_SIEM_HOST set to your collector. |
Generic syslog output. No product-specific app, parser or dashboard is published for any SIEM. | Limited | SIEM export |
| Microsoft Sentinel (native connector) | Export or enforcement destination | The ClawMetry daemon sends to your Sentinel workspace through the Azure Monitor Logs Ingestion API | One table, ClawMetryGovernance_CL: Guard incidents, decided pre-tool gate approvals, Guard policy decisions and audit log entries, with free text redacted before it leaves the node. Each record lists the Compliance Pack controls it is evidence for. |
Enterprise plan and the Pro extension 0.7.28 or later. A Microsoft Entra application, a table and data collection rule in your workspace, the CLAWMETRY_SENTINEL_* settings, and HTTPS endpoints. |
It has not yet been run against a live Azure tenant by us. Records do not carry the OWASP or MITRE ATLAS references Guard findings have, and policy decisions map to no control yet. The Azure template is provided on request, not in the installed package. | Limited | Microsoft Sentinel |
| MCP server for your agents | Export or enforcement destination | ClawMetry answers your agent over stdio on the same machine | Session lists, cost summaries, session traces, events, incidents, per-session Guard status, signal rates, operator notes and ClawMetry's own health. | The open-source package, a running ClawMetry daemon, and an MCP client such as Claude Code or Cursor. | Local over stdio. No tool acts on a process; one tool lets the agent file a note for the operator. It gives the agent the content of past sessions, so attach it deliberately. | Available | MCP overview |
| CI pipelines: eval suites as a merge gate | Export or enforcement destination | Your CI job runs clawmetry eval and reads its exit code |
Pass or fail per golden test, with a JSON report. | Pro plan, and an API key for the judge model. | Gates on eval results only. | Available | clawmetry eval |
| Pull request provenance in CI: agent sessions on changed files, as SARIF, a comment and an opt-in gate | Export or enforcement destination | Your CI job runs clawmetry trace report, or the GitHub Action, on a bundle exported from the developer's machine |
Which agent sessions wrote which changed files, with runtime, model, cost and Guard findings; your own scanners' SARIF findings on those files with the session attached; one pull request comment; a recorded gate decision. | clawmetry 0.12.877 or later on the developer's machine to export the bundle, and the GitHub Action or the same command in CI. Uploading SARIF to GitHub code scanning needs the security-events: write permission. |
Association, not proof of cause: each link carries its basis, and a later human edit is not excluded. It is not a code scanner. The gate is off unless you enable it. The SARIF upload and a live pull request comment have not yet been verified against github.com, and pull requests from forks get a read-only token. No Azure DevOps task. | Limited | PR provenance |