Model Context Protocol (MCP) Server for Error Monitoring and AI Observability

sentry · Tech · 2025 · source Agentic: yes · promotional bias: med · source-backed · extracted with prompt 0.5.6 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: Error monitoring and AI observability for developer agents

Sentry wanted LLM-powered development tools to access current error, project, performance, and root-cause context from Sentry without forcing users to run brittle local connectors or manually stitch together API calls.

Constraints:

Architecture

Sentry built a hosted MCP server at https://mcp.sentry.dev/mcp that exposes Sentry monitoring context to MCP-compatible clients. The service supports hosted remote mode and local STDIO compatibility, OAuth login through existing Sentry organizations, Streamable HTTP with fallback to SSE, and 16 tool calls for project information, issue data, project/DSN creation, and Seer-powered root-cause analysis. Cloudflare Workers provide the edge deployment target, Durable Objects maintain stateful user-session storage, and Cloudflare's OAuth provider library plugs into Sentry authentication. Sentry also added MCP-specific JavaScript SDK instrumentation to observe tool calls, protocol behavior, Durable Objects, and the server itself.

Diagram confidence: explicit (topology described by the source).

flowchart LR
  Client["MCP-capable client"] --> Server["Sentry MCP server"]
  Client --> LLM["LLM"]
  Server --> OAuth["OAuth integration"]
  Server --> Workers["Cloudflare Workers"]
  Workers --> Durable["Durable Objects"]
  Server --> API["Sentry API"]
  API --> Seer["Seer agent"]
  Server --> Seer
  SDK["JavaScript SDK instrumentation"] --> Server
Drag to pan · scroll or use +/− to zoom · reconstructed from the source's described component topology

Components

Component Role Interacts with
MCP-capable client Editor, LLM provider, or AI development environment that connects to Sentry MCP and exposes tools to the model. Sentry MCP server, LLM
Sentry MCP server Hosted or local MCP server exposing Sentry context and tool calls. MCP-capable client, OAuth integration, Cloudflare Workers, Sentry API, Seer agent
OAuth integration Authentication path that logs users in through existing Sentry organizations and avoids manual API tokens. Sentry MCP server, Sentry API
Cloudflare Workers Global edge deployment target for the hosted MCP workload. Sentry MCP server, Durable Objects
Durable Objects Persistent stateful storage between user sessions and context interactions. Cloudflare Workers, Sentry MCP server
Sentry API Backend API for project information, issue details, DSNs, and monitoring data. Sentry MCP server, Seer agent
Seer agent Sentry's specialized AI agent for root-cause analysis and issue-fix operations. Sentry MCP server, Sentry API
JavaScript SDK instrumentation SDK support for observing MCP protocol behavior, tool calls, and server infrastructure. Sentry MCP server
LLM Model that reads MCP tool descriptions and chooses project, issue, and Seer tool calls. MCP-capable client

Data flow

A developer configures the Sentry MCP endpoint in an MCP-capable editor or LLM client. The client prompts the user through OAuth, then exposes Sentry tool names, descriptions, schemas, and parameters to the LLM. The LLM can call tools such as project lookup, issue retrieval, or Seer issue-fix operations. Tool results flow back into the LLM context, and clients may store stable details such as default organization names to avoid repeated calls. Sentry monitors the MCP server, tool-call performance, infrastructure behavior, and Durable Object state through its own instrumentation.

3 supporting quotes
“The Sentry MCP can be added to most providers that support OAuth by adding the following content to your MCP Configuration file:”article, stated
“Without MCP, the LLM would have no knowledge available on what projects were available for the user. MCP ends up becoming that bridge”article, stated
“It can then chain the different tool calls together to kick off Seer’s analysis run, and return the RCA when it’s ready.”article, stated

Why this architecture

Sentry chose a hosted remote MCP server because local STDIO setup had too many operational and security sharp edges for broad adoption. OAuth lets Sentry reuse organization access controls instead of pushing long-lived API tokens through client configuration. Cloudflare Workers and Durable Objects provide a fast path to production-scale hosted operation, while Streamable HTTP plus SSE fallback keeps client compatibility. The MCP layer complements Seer by supplying context and triggering specialized root-cause analysis rather than trying to replace Sentry's purpose-built debugging agent.

3 supporting quotes
“It’s lower friction and you’ll always have the latest functionality/stability”article, stated
“Support for OAuth was necessary, and let’s treat it like another application connection coming into Sentry. We don’t need to reinvent the wheel”article, stated
“Cloudflare workers gave us the scale of the global edge network, and an easy deployment target to run the workload on”article, stated

Models used: large language models · Seer · AI agents · LLMs in editors

Scale

Alternatives considered (roads not taken)

Use only local STDIO MCP servers

Rejected because: Local setup required cloning repositories, editing configs, tracking file paths, updating servers, and managing tokens.

“works fine for advanced users, but cloning the repo, setting up the configurations, remembering where the file path is”article, stated
“cloning the repo, setting up the configurations, remembering where the file path is, and continuing to keep it updated is a lot of sharp edges”article, stated

Require manually generated API tokens

Rejected because: Sentry considered token passing not great, and OAuth could reuse existing application-connection patterns and organization permissions.

“Making people create API tokens and pass them around is… not great.”article, stated
“to enable logging in via your existing Sentry organization, so you’ll have clear access”article, stated

Use MCP itself as the deep debugging agent

Rejected because: Sentry says MCP brings context into an LLM, while Seer is purpose-built for root-cause analysis and issue fixing.

“MCP in general is not built for debugging - it’s built for bringing context into your LLM from external sources.”article, stated
“for understanding why your applications are breaking, and how you can get them fixed.”article, stated

Support only Streamable HTTP

Rejected because: Some clients do not support Streamable HTTP yet, so Sentry automatically falls back to SSE.

“if your client doesn’t support it”article, stated
“If your client doesn’t support Streamable HTTP yet, this will automatically fallback to the SSE connection.”article, stated

Patterns

tool-use-orchestration (architecture)

Sentry exposes project, issue, DSN, and Seer operations as MCP tools that an LLM can chain together for debugging workflows.

“including prompts, to support bringing wide context in from Sentry”article, stated
“these tool calls are chained together. Running Seer’s Issue Fix against a bug that appeared actually requires several tool calls”article, stated

observability-tracing (operations)

The MCP server is both an observability interface for Sentry data and an instrumented production service monitored through Sentry's SDK and platform.

“which gives a lot more visibility to the underlying functionality of the protocol and tool calls.”article, stated
“monitoring the core MCP functionality to monitoring the Durable Object layer as well.”article, stated

safety-policy-enforcement (security)

Sentry uses OAuth and organization-scoped access instead of manual API token passing, making MCP connections behave like normal Sentry application integrations.

“to enable logging in via your existing Sentry organization, so you’ll have clear access”article, stated
“treat it like another application connection coming into Sentry”article, stated

durable-execution (infrastructure)

Cloudflare Durable Objects keep state between user sessions, supporting a remote MCP service that needs persistent auth/session context.

“provided persistent, stateful storage between user sessions”article, stated
“This combination gave us a quick path to getting the service out in the wild for users to consume”article, stated

context-engineering (context)

Sentry uses MCP as a context bridge so LLMs can ground debugging answers in current Sentry projects, issues, and tool results rather than relying on stale model memory.

“Using MCP as a bridge to give users access to that context from within the LLM is clearly a good way to enable getting that.”article, stated
“so the LLM doesn’t assume (aka hallucinate) context on its own.”article, stated

Tradeoffs

Prefer hosted remote MCP.

Chose Remote hosted Sentry MCP service over Local STDIO-only server.

Rationale: Hosted MCP reduces setup friction, keeps users on the latest functionality, and fits Sentry's scale requirements.

“we strongly recommend the hosted version. It’s lower friction and you’ll always have the latest functionality/stability”article, stated
“It’s lower friction and you’ll always have the latest functionality/stability”article, stated

Authenticate through OAuth.

Chose Existing Sentry organization OAuth flow over Manual API token creation and passing.

Rationale: OAuth gives users clear access to organizations and projects while reusing Sentry's application-connection model.

“Making people create API tokens and pass them around is… not great.”article, stated
“to enable logging in via your existing Sentry organization, so you’ll have clear access”article, stated

Separate MCP context access from Seer root-cause analysis.

Chose MCP as context/access layer plus Seer as specialized AI debugger over A single general MCP-only debugging flow.

Rationale: MCP is best at bringing context into LLMs; Seer is optimized for deep bug and performance issue analysis.

“MCP in general is not built for debugging - it’s built for bringing context into your LLM from external sources.”article, stated
“At the end of the day, they are different tools for different problems, in a similar space.”article, stated

Gotchas (what bit them)

Local MCP installation creates sharp edges for normal users.

Root cause: Users must clone code, configure paths, update dependencies, and manage tokens just to connect a service.

Resolution: Provide a hosted remote MCP endpoint and recommend it as the preferred path.

“cloning the repo, setting up the configurations, remembering where the file path is, and continuing to keep it updated is a lot of sharp edges”article, stated
“We knew early on that our MCP was going to need to be a hosted / remote solution.”article, stated

MCP tool chaining can be inefficient or suboptimal.

Root cause: Declarative tool selection and pathing between tool chains are still early, and complex debugging may require many intermediate calls.

Resolution: Keep Seer as the purpose-built debugging agent and use MCP to supply context or trigger Seer rather than manually reconstructing root-cause workflows.

“controlling pathing between different till chaining, and even parts of authentication still have their share of bugs.”article, stated
“Trying to accomplish the same thing using MCP would result in you manually stitching together many different tool calls”article, stated

Outdated model knowledge produces bad setup and debugging advice.

Root cause: Training data may contain old SDK versions or stale API behavior.

Resolution: Use MCP tools to retrieve current Sentry project and issue context directly from the platform.

“the models are forced to rely on only the content they’ve been trained on - which often ends up being out of date”article, stated
“remember that time you tried to have an LLM setup Sentry and it installed that weirdly ancient version of the React SDK?”article, stated

Evaluation

Sentry evaluates the MCP server by monitoring tool calls, protocol behavior, Durable Object behavior, usage, errors, response characteristics, and whether AI clients can chain tools into useful issue-analysis workflows.

Metrics: tool-call visibility · MCP protocol errors · Durable Object health · tool-call chaining success · OAuth connection success · response latency · usage patterns · Seer RCA readiness

3 supporting quotes
“gives a lot more visibility to the underlying functionality of the protocol and tool calls.”article, stated
“monitoring the core MCP functionality to monitoring the Durable Object layer as well.”article, stated
“helps Sentry understand how developers are using different tools and identify opportunities for optimization or new tool development.”summary, implied

Outcomes

The implementation is production-oriented rather than a toy server: Sentry highlights hosted remote operation, significant user load, global edge deployment, OAuth, 16 tool calls, and monitoring of the MCP server itself.

3 supporting quotes
“yes, as the title of this post suggests - Sentry has an MCP server.”article, stated
“at production scale, quickly, pretty easy”article, stated
“the MCP can (as we discussed earlier) make calls to Seer to kick off the issue fix operations.”article, stated