I the possibility to create an MCP server for an observability utility in an effort to present the AI agent with dynamic code evaluation capabilities. Due to its potential to rework functions, MCP is a know-how I’m much more ecstatic about than I initially was about genAI normally. I wrote extra about that and a few intro to MCPs normally in a earlier post.
Whereas an preliminary POCs demonstrated that there was an immense potential for this to be a power multiplier to our product’s worth, it took a number of iterations and a number of other stumbles to ship on that promise. On this submit, I’ll attempt to seize among the classes discovered, as I believe that this could profit different MCP server builders.
My Stack
- I used to be utilizing Cursor and vscode intermittently as the primary MCP shopper
- To develop the MCP server itself, I used the .NET MCP SDK, as I made a decision to host the server on one other service written in .NET
Lesson 1: Don’t dump your whole knowledge on the agent
In my utility, one software returns aggregated data on errors and exceptions. The API may be very detailed because it serves a posh UI view, and spews out giant quantities of deeply linked knowledge:
- Error frames
- Affected endpoints
- Stack traces
- Precedence and tendencies
- Histograms
My first hunch was to easily expose the API as is as an MCP software. In any case, the agent ought to be capable to make extra sense of it than any UI view, and catch on to attention-grabbing particulars or connections between occasions. There have been a number of situations I had in thoughts as to how I might count on this knowledge to be helpful. The agent may mechanically provide fixes for current exceptions recorded in manufacturing or within the testing setting, let me learn about errors that stand out, or assist me deal with some systematic issues which might be the underlying root reason for the problems.
The fundamental premise was subsequently to permit the agent to work its ‘magic’, with extra knowledge probably that means extra hooks for the agent to latch on in its investigation efforts. I shortly coded a wrapper round our API on the MCP endpoint and determined to begin with a primary immediate to see whether or not every thing is working:
We will see the agent was sensible sufficient to know that it wanted to name one other software to seize the setting ID for that ‘check’ setting I discussed. With that at hand, after discovering that there was truly no current exception within the final 24 hours, it then took the freedom to scan a extra prolonged time interval, and that is when issues bought a bit bizarre:
What an odd response. The agent queries for exceptions from the final seven days, will get again some tangible outcomes this time, and but proceeds to ramble on as if ignoring the info altogether. It continues to try to use the software in numerous methods and completely different parameter combos, clearly fumbling, till I discover it flat out calls out the truth that the info is totally invisible to it. Whereas errors are being despatched again within the response, the agent truly claims there are no errors. What’s going on?
After some investigation, the issue was revealed to be the truth that we’ve merely reached a cap within the agent’s capability to course of giant quantities of knowledge within the response.
I used an current API that was extraordinarily verbose, which I initially even thought of to be a bonus. The top end result, nonetheless, was that I by some means managed to overwhelm the mannequin. Total, there have been round 360k characters and 16k phrases within the response JSON. This consists of name stacks, error frames, and references. This ought to have been supported simply by trying on the context window restrict for the mannequin I used to be utilizing (Claude 3.7 Sonnet ought to help as much as 200k tokens), however however the big knowledge dump left the agent totally stumped.
One technique can be to alter the mannequin to at least one that helps a good greater context window. I converted to the Gemini 2.5 professional mannequin simply to check that concept out, because it boasts an outrageous restrict of 1 million tokens. Positive sufficient, the identical question now yielded a way more clever response:
That is nice! The agent was in a position to parse the errors and discover the systematic reason for lots of them with some primary reasoning. Nevertheless, we will’t depend on the person utilizing a particular mannequin, and to complicate issues, this was output from a comparatively low bandwidth testing setting. What if the dataset had been even bigger?
To unravel this concern, I made some basic modifications to how the API was structured:
- Nested knowledge hierarchy: Preserve the preliminary response targeted on high-level particulars and aggregations. Create a separate API to retrieve the decision stacks of particular frames as wanted.
- Improve queryability: The entire queries made to this point by the agent used a really small web page dimension for the info (10), if we would like the agent to have the ability to to entry extra related subsets of the info to suit with the constraints of its context, we have to present extra APIs to question errors based mostly on completely different dimensions, for instance: affected strategies, error kind, precedence and influence and many others.
With the brand new modifications, the software now constantly analyzes necessary new exceptions and comes up with repair ideas. Nevertheless, I glanced over one other minor element I wanted to type earlier than I may actually use it reliably.
Lesson 2: What’s the time?
The keen-eyed reader might have seen that within the earlier instance, to retrieve the errors in a particular time vary, the agent makes use of the ISO 8601 time period format as a substitute of the particular dates and instances. So as a substitute of together with normal ‘From’ and ‘To’ parameters with datetime values, the AI despatched a period worth, for instance, seven days or P7D, to point it desires to test for errors up to now week.
The explanation for that is considerably unusual — the agent may not know the present date and time! You may confirm that your self by asking the agent that easy query. The under would have made sense had been it not for the truth that I typed that immediate in at round midday on Might 4th…
Utilizing time period values turned out to be an ideal resolution that the agent dealt with fairly properly. Don’t neglect to doc the anticipated worth and instance syntax within the software parameter description, although!
Lesson 3: When the agent makes a mistake, present it find out how to do higher
Within the first instance, I used to be truly bowled over by how the agent was in a position to decipher the dependencies between the completely different software calls In an effort to present the suitable setting identifier. In finding out the MCP contract, it discovered that it needed to name on a dependent one other software to get the checklist of setting IDs first.
Nevertheless, responding to different requests, the agent would generally take the setting names talked about within the immediate verbatim. For instance, I seen that in response to this query: evaluate gradual traces for this methodology between the check and prod environments, are there any vital variations? Relying on the context, the agent would generally use the setting names talked about within the request and would ship the strings “check” and “prod” because the setting ID.
In my authentic implementation, my MCP server would silently fail on this state of affairs, returning an empty response. The agent, upon receiving no knowledge or a generic error, would merely stop and attempt to clear up the request utilizing one other technique. To offset that conduct, I shortly modified my implementation in order that if an incorrect worth was offered, the JSON response would describe precisely what went flawed, and even present a sound checklist of attainable values to save lots of the agent one other software name.
This was sufficient for the agent, studying from its mistake, it repeated the decision with the right worth and by some means additionally prevented making that very same error sooner or later.
Lesson 4: Deal with person intent and never performance
Whereas it’s tempting to easily describe what the API is doing, generally the generic phrases don’t fairly enable the agent to comprehend the kind of necessities for which this performance may apply greatest.
Let’s take a easy instance: My MCP server has a software that, for every methodology, endpoint, or code location, can point out the way it’s getting used at runtime. Particularly, it makes use of the tracing knowledge to point which utility flows attain the particular operate or methodology.
The unique documentation merely described this performance:
[McpServerTool,
Description(
@"For this method, see which runtime flows in the application
(including other microservices and code not in this project)
use this function or method.
This data is based on analyzing distributed tracing.")]
public static async Activity GetUsagesForMethod(IMcpService shopper,
[Description("The environment id to check for usages")]
string environmentId,
[Description("The name of the class. Provide only the class name without the namespace prefix.")]
string codeClass,
[Description("The name of the method to check, must specify a specific method to check")]
string codeMethod) The above represents a functionally correct description of what this software does, nevertheless it doesn’t essentially make it clear what forms of actions it is perhaps related for. After seeing that the agent wasn’t selecting this software up for numerous prompts I assumed it might be pretty helpful for, I made a decision to rewrite the software description, this time emphasizing the use instances:
[McpServerTool,
Description(
@"Find out what is the how a specific code location is being used and by
which other services/code.
Useful in order to detect possible breaking changes, to check whether
the generated code will fit the current usages,
to generate tests based on the runtime usage of this method,
or to check for related issues on the endpoints triggering this code
after any change to ensure it didnt impact it"Updating the text helped the agent realize why the information was useful. For example, before making this change, the agent would not even trigger the tool in response to a prompt similar to the one below. Now, it has become completely seamless, without the user having to directly mention that this tool should be used:
Lesson 5: Document your JSON responses
The JSON standard, at least officially, does not support comments. That means that if the JSON is all the agent has to go on, it might be missing some clues about the context of the data you’re returning. For example, in my aggregated error response, I returned the following score object:
"Score": {"Score":21,
"ScoreParams":{ "Occurrences":1,
"Trend":0,
"Recent":20,
"Unhandled":0,
"Unexpected":0}}Without proper documentation, any non-clairvoyant agent would be hard pressed to make sense of what these numbers mean. Thankfully, it is easy to add a comment element at the beginning of the JSON file with additional information about the data provided:
"_comment": "Each error contains a link to the error trace,
which can be retrieved using the GetTrace tool,
information about the affected endpoints the code and the
relevant stacktrace.
Each error in the list represents numerous instances
of the same error and is given a score after its been
prioritized.
The score reflects the criticality of the error.
The number is between 0 and 100 and is comprised of several
parameters, each can contribute to the error criticality,
all are normalized in relation to the system
and the other methods.
The score parameters value represents its contributation to the
overall score, they include:
1. 'Occurrences', representing the number of instances of this error
compared to others.
2. 'Trend' whether this error is escalating in its
frequency.
3. 'Unhandled' represents whether this error is caught
internally or poropagates all the way
out of the endpoint scope
4. 'Unexpected' are errors that are in high probability
bugs, for example NullPointerExcetion or
KeyNotFound",
"EnvironmentErrors":[]This permits the agent to clarify to the person what the rating means in the event that they ask, but in addition feed this rationalization into its personal reasoning and proposals.
Selecting the best structure: SSE vs STDIO,
There are two architectures you should utilize in creating an MCP server. The extra widespread and broadly supported implementation is making your server accessible as a command triggered by the MCP shopper. This may very well be any CLI-triggered command; npx, docker, and python are some widespread examples. On this configuration, all communication is finished through the method STDIO, and the method itself is operating on the shopper machine. The shopper is accountable for instantiating and sustaining the lifecycle of the MCP server.
This client-side structure has one main disadvantage from my perspective: Because the MCP server implementation is run by the shopper on the native machine, it’s a lot tougher to roll out updates or new capabilities. Even when that drawback is by some means solved, the tight coupling between the MCP server and the backend APIs it is dependent upon in our functions would additional complicate this mannequin by way of versioning and ahead/backward compatibility.
For these causes, I selected the second kind of MCP Server — an SSE Server hosted as part of our utility companies. This removes any friction from operating CLI instructions on the shopper machine, in addition to permits me to replace and model the MCP server code together with the applying code that it consumes. On this state of affairs, the shopper is supplied with a URL of the SSE endpoint with which it interacts. Whereas not all shoppers at present help this feature, there’s a good commandMCP referred to as supergateway that can be utilized as a proxy to the SSE server implementation. Meaning customers can nonetheless add the extra broadly supported STDIO variant and nonetheless eat the performance hosted in your SSE backend.
MCPs are nonetheless new
There are various extra classes and nuances to utilizing this deceptively easy know-how. I’ve discovered that there’s a huge hole between implementing a workable MCP to at least one that may truly combine with person wants and utilization situations, even past these you will have anticipated. Hopefully, because the know-how matures, we’ll see extra posts on Best Practices.
Need to Join? You may attain me on Twitter at @doppleware or through LinkedIn.
Comply with my mcp for dynamic code evaluation utilizing observability at https://github.com/digma-ai/digma-mcp-server
