任意のツール呼び出しエージェント向けの、ドロップインMCPサーバー + 実行メトリクス記録ツール。エージェントのループで、実行後に record_session(prompt, model_id, loop_result) を指すだけで、以下の機能が無料で得られます。
実際のMCPサーバー（Streamable HTTP）— Claude Desktop、Cursor、独自のチャットUIなど、任意のMCPクライアントが接続可能。セッション履歴、コスト、トークン/ツールメトリクスを公開する7つの読み取り専用ツールを提供し、さらに...
コンテキストウィンドウエクスプローラー — モデルのコンテキストウィンドウに何が入力されたかを、ブロックごとに完全に透明化。正直な（明示的に「推定値」とラベル付けされた）トークン数、比例セグメントバー、ブロックごとのクリック拡張可能な詳細パネルを備えています。
*(このパッケージと同時に開発されたリファレンスチャットUI — sre-investigation-agent — のスクリーンショット。上記のパネルは、任意のMCPクライアントが get_context_timeline をクエリした際に取得できるものです。) *
ほとんどのエージェント観測ツールは、自身のUIが既に表示しているデータを再表示します。このツールは、通常はまったく見えないものを表示します。システムプロンプト vs. ツール仕様 vs. 推論 vs. ツール呼び出し/結果 vs. 最終回答、それらが実際にコンテキストに入力された順序、モデルの実際のコンテキストウィンドウに対する累計トークン数、そしてそれらのブロックのうちエンドユーザーに表示されるものと表示されないオーバーヘッドを明らかにします。
uv add mcp-context-inspector # or: pip install mcp-context-inspector
# while co-developing locally against an editable checkout:
uv add --editable ../mcp-context-inspectorRelated MCP server: ai-usage-metrics-mcp
from metrics import store
session_id = store.record_session(prompt, model_id, loop_result)loop_result はエージェントループが返す任意のものです。このパッケージが必要とする形式は以下の通りです。
{
"trace": [{"tool": "...", "args": {...}, "status": "ok"}, ...],
"turns": [{"input_tokens": int, "output_tokens": int, "latency_ms": int}, ...],
"input_tokens": int, "output_tokens": int, "total_tokens": int, "latency_ms": int,
"context_blocks": [ # optional — omit and you just lose the Explorer, nothing crashes
{"category": "system", "label": "...", "char_count": int, "token_estimate": int, "turn_n": int | None},
...
],
}context_blocks のカテゴリ: system, tools, user, reasoning, thinking, tool_call, tool_result （オプションで失敗を色分けするための "status" キーを含む）、answer。
uv run python -m mcp_server.serverMCP_AUTH_TOKEN が設定されていない場合、起動時に生成して出力します。Jupyterサーバーが出力するトークンと同じ信頼モデルです。再起動後も安定した値が必要な場合は、自分で設定してください。任意のMCPクライアントを http://127.0.0.1:8787/mcp に Authorization: Bearer <token> ヘッダー付きで接続します。
STORAGE_BACKEND=sqlite（デフォルト、ローカル開発 — data/metrics.db）または STORAGE_BACKEND=dynamodb（METRICS_TABLE/AWS_REGION を設定）— どちらの場合も同じ関数シグネチャで、呼び出し側はどちらがアクティブかを意識しません。
get_session_metrics, get_token_breakdown, get_tool_metrics, get_agent_trace, get_cost_estimate, get_recent_sessions, get_context_timeline。プレーンなRESTエンドポイントも /api/* で公開されています — curlに便利なデバッグ代替手段で、同じ内部の metrics/store.py の関数を呼び出します。
sre-investigation-agent —
このパッケージの基盤となり、同時に開発されたリファレンスチャットUI + Bedrockエージェント。
The ordered sequence of tool calls for one session -- each entry {name, args, status}.
Use to replay what the agent actually did, in execution order; use get_tool_metrics for aggregate counts.
session_id: id from get_recent_sessions / record_session.
Returns [] for an unknown or non-owned session.
| Name | Required | Description | Default |
|---|---|---|---|
| session_id | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint=true, idempotentHint=true, and destructiveHint=false. The description adds meaningful behavior beyond that: the ordered return shape (each entry with name, args, status) and the empty-array return for unknown or non-owned sessions, which is not implied by the schema.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Three sentences, each purpose-specific: output definition, usage vs. alternative, and parameter sourcing. No filler and the most important information is front-loaded.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a single-parameter read-only tool with no output schema, the description provides the output format, ordering guarantee, empty-array edge case, and parameter provenance. Nothing essential is missing for correct invocation.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 0%, so the description compensates by documenting session_id as 'id from get_recent_sessions / record_session'. This tells the agent where to obtain the ID and what it refers to, which the bare string type does not convey.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
Description opens with 'The ordered sequence of tool calls for one session -- each entry {name, args, status}', which clearly states the verb, resource, and output structure. It explicitly contrasts with get_tool_metrics, so an agent can tell them apart.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Description explicitly instructs: 'Use to replay what the agent actually did, in execution order; use get_tool_metrics for aggregate counts.' This names the alternative and the condition selecting it, leaving no ambiguity about when to use this tool.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Ordered, categorized breakdown of everything that entered ONE session's context window.
Each block (system prompt, tool specs, injected context, user turns, reasoning, tool calls/results, final answer) is marked user-visible vs invisible overhead, with cumulative character-based token estimates against the model's real window size. Use for context-window composition analysis; use get_session_metrics for exact provider token usage. The estimates here are character-based, not exact Bedrock counts.
session_id: id from get_recent_sessions / record_session.
Returns [] for an unknown or non-owned session, or one recorded without the optional context_blocks field.
| Name | Required | Description | Default |
|---|---|---|---|
| session_id | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Beyond the annotations (readOnlyHint, idempotentHint, destructiveHint), the description discloses that estimates are character-based rather than exact Bedrock counts, that it returns [] for unknown/non-owned sessions or sessions lacking the optional context_blocks field, and that blocks are classified as user-visible vs invisible overhead. These are non-obvious behaviors an agent would not know from the schema alone.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Every sentence earns its place: the first sentence states the core purpose, the second explains the breakdown format, the third gives usage guidance, and the final two cover parameters and return behavior. There is no filler, and the key differentiator is front-loaded.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a tool with one string parameter, read-only annotations, and no output schema, the description provides sufficient context: what the tool returns, how it behaves for missing/unknown sessions, what the session_id should be, and how the estimates differ from exact metrics. An agent can both select and safely invoke this tool without further assumptions.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The schema only defines session_id with no textual description (0% coverage), but the description explicitly says 'session_id: id from get_recent_sessions / record_session.' This completely compensates for the schema gap by telling the agent exactly where the id comes from.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description opens with a specific, differentiated statement: 'Ordered, categorized breakdown of everything that entered ONE session's context window.' It names the resource (a session's context window) and the operation's output form, and it later explicitly distinguishes itself from get_session_metrics by noting exact token usage is not what this returns.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description explicitly states when to use the tool: 'Use for context-window composition analysis.' It also names the alternative for exact token usage: 'use get_session_metrics for exact provider token usage.' This gives the agent an unambiguous routing decision.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Estimated USD cost as a float.
Pass session_id for one session's cost, or period_seconds for the summed cost of your sessions in the last N seconds. Give exactly one; non-owners are scoped to their own sessions. These are estimates from token counts and a static price table, not billed amounts.
session_id: optional session id. period_seconds: optional lookback window in seconds (e.g. 86400 for the last day).
Returns 0.0 for an unknown or non-owned session_id, and for a period with no matching sessions.
| Name | Required | Description | Default |
|---|---|---|---|
| session_id | No | ||
| period_seconds | No |
| Name | Required | Description |
|---|---|---|
| result | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already mark this as read-only, idempotent, and non-destructive, so the bar is lower. The description adds meaningful behavioral context: estimates are not billed amounts, and unknown/non-owned session IDs as well as empty periods return 0.0. These are exactly the edge cases an agent needs to anticipate.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is compact and front-loaded with the core meaning, then moves to usage modes, then edge cases. Each sentence earns its place; no filler or repetition of annotation values.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the annotations already cover safety and idempotency, and an output schema exists, the description supplies all remaining needed context: parameter semantics, exclusivity, scoping, estimate nature, and zero-value fallback behavior. An agent can invoke this tool correctly without further inference.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 0%, so the description carries the full burden for parameter meaning. It explains both parameters in plain language, gives an example value for period_seconds, and clarifies mutual exclusivity plus ownership scoping. This fully compensates for the bare schema.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description opens with a precise statement of purpose: 'Estimated USD cost as a float.' It clearly identifies the resource (session cost) and the computation basis (token counts and a static price table), and distinguishes this from billed amounts, which separates it from related metrics tools.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description gives explicit usage guidance: pass exactly one of session_id or period_seconds, explains what each mode computes, and notes non-owner scoping. It does not explicitly name sibling tools as alternatives, but for this tool the parameter-mode distinction is the main usage decision and it is covered clearly.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
List recent sessions, newest first: [{session_id, prompt, model_id, created_at, ...}].
Call this first to discover session_ids for the other get_* tools. Non-owners see only their own sessions; the owner token sees all.
limit: max rows to return (default 10), newest first.
| Name | Required | Description | Default |
|---|---|---|---|
| limit | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnly, idempotent, and non-destructive behavior, so the description does not need to restate those. It adds meaningful behavioral context by explaining ownership-based visibility: non-owners see only their own sessions while the owner token sees all, plus the output format in the sample.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is compact and front-loaded: a one-line summary, a usage-direction sentence, and a parameter note. Every sentence serves a distinct purpose with no filler or repetition.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
This is a low-complexity tool with one optional parameter. The description covers what it returns, how results are ordered, the visibility rules, and how it relates to sibling tools. There is no output schema, but the sample shape compensates adequately.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 0%, so the description carries full responsibility for explaining the 'limit' parameter. It does so clearly: 'max rows to return (default 10), newest first.' This adds semantic meaning beyond the raw integer schema and fully compensates for the missing schema description.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description opens with a specific verb and resource: 'List recent sessions, newest first' and includes a sample result shape. It also distinguishes this tool's role as the discovery entry point for session_ids used by the other get_* tools, making it easy for an agent to know exactly what it does.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The phrase 'Call this first to discover session_ids for the other get_* tools' gives explicit sequencing and intent, which is strong usage guidance. It does not spell out exclusion cases, but the context makes it clear this is the intended first step before sibling get_* tools.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Full metrics for ONE recorded session: metadata plus per-prompt tokens, latency, and cost.
Use when you have a session_id and need exact provider usage numbers for that session. For the block-by-block token composition of the context window use get_context_timeline instead; for a cost total across many sessions use get_cost_estimate.
session_id: the opaque, case-sensitive id returned by record_session or listed by get_recent_sessions.
Returns {"error": "session not found"} if the id is unknown or not owned by the caller (the two are deliberately indistinguishable).
| Name | Required | Description | Default |
|---|---|---|---|
| session_id | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint, idempotentHint, and non-destructive behavior. The description adds valuable behavior beyond that: it specifies the exact error response format and discloses that unknown and unowned IDs are deliberately indistinguishable, which is useful privacy/security context an agent needs to interpret failures correctly.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is compact yet information-dense. It front-loads the core purpose, then covers usage alternatives, parameter semantics, and error behavior in ordered paragraphs. Every sentence adds necessary information without redundancy.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Despite having no output schema, the description adequately summarizes the return content: metadata plus per-prompt tokens, latency, and cost. It also explains error behavior and how to obtain a valid session_id. For a single-parameter, read-only tool, nothing essential is missing for an agent to select and invoke it correctly.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 0%, so the description carries the full burden for the session_id parameter. It does so thoroughly: 'the opaque, case-sensitive id returned by record_session or listed by get_recent_sessions' adds provenance and case-sensitivity details that the bare schema string type cannot convey.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description uses a specific verb and resource: 'Full metrics for ONE recorded session: metadata plus per-prompt tokens, latency, and cost.' It names the exact scope (one session) and distinguishes itself from siblings like get_context_timeline and get_cost_estimate without needing to inspect those tools.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
It explicitly states when to use it: 'Use when you have a session_id and need exact provider usage numbers for that session.' It then names alternatives for other cases (get_context_timeline for block-by-block context composition, get_cost_estimate for cross-session totals), providing clear selection criteria.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Per-turn input/output token and latency breakdown for one session, in turn order.
Use to see how token usage grew turn-by-turn within a session; use get_session_metrics for session totals.
session_id: id from get_recent_sessions / record_session.
Returns [] for an unknown or non-owned session.
| Name | Required | Description | Default |
|---|---|---|---|
| session_id | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already communicate read-only, idempotent, and non-destructive behavior. The description adds useful behavioral details beyond the annotations: results are in turn order, and an unknown or non-owned session returns an empty list. It does not specify units for latency, but the annotation coverage lowers the burden.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is compact and well-structured: a definitional first sentence, a usage-routing second sentence, and a parameter/edge-case note. Every sentence earns its place without redundancy.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a single-parameter read-only operation, the description covers purpose, expected output shape, turn ordering, token and latency aspects, parameter provenance, and the empty-list edge case. No output schema exists, but the description provides enough conceptual shape to invoke the tool correctly.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The input schema provides no description for session_id, so the description compensates by explaining that session_id comes from get_recent_sessions or record_session. This gives an agent the source and provenance of the parameter, which is meaningful given the 0% schema description coverage.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description states a specific verb and resource: it provides per-turn input/output token and latency breakdown for one session, in turn order. It also explicitly distinguishes itself from get_session_metrics, so an agent can tell when this tool applies.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description gives direct usage context: use this tool to see how token usage grew turn-by-turn within a session, and use get_session_metrics for session totals. This explicit routing to a sibling tool makes the selection criteria clear.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Tool-call counts grouped by status (ok / error), for one session or aggregated.
Use for a quick success/failure summary; use get_agent_trace for the ordered call list. Non-owners only ever see their own sessions; the owner token sees everyone's.
session_id: optional. Omit for the aggregate across all your sessions; pass an id for just that session.
| Name | Required | Description | Default |
|---|---|---|---|
| session_id | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint, idempotentHint, and non-destructive behavior. The description adds valuable behavioral context: results are grouped by ok/error, omitting session_id aggregates across sessions, and non-owners only see their own sessions while the owner sees everyone's.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is compact and well-structured: core purpose first, then usage guidance, ownership caveat, and parameter detail. Every sentence earns its place with no redundancy.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a simple one-parameter read-only metrics tool with no output schema, the description covers purpose, when to use an alternative, ownership scoping, and parameter semantics. Nothing essential is missing.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 0%, but the description fully compensates by explaining the only parameter, session_id: omit for the aggregate across all sessions, or pass an id for just that session. This is exactly the semantic information an agent needs.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
States a specific verb and resource: it returns tool-call counts grouped by status (ok/error) for one session or aggregated. It distinguishes itself from get_agent_trace by framing itself as a summary rather than an ordered call list.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Explicitly says to use this tool for a quick success/failure summary and to use get_agent_trace for the ordered call list. It also gives clear direction on omitting versus passing session_id, plus an ownership-based visibility caveat.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Append ONE agent execution's metrics to this server's store; returns the new session_id.
This is how a remote agent gets its own runs into the server, rather than only being able to query what the server owner recorded locally. Attributed to the connected identity: your Google account if you signed in via /auth/login, or owner=None for the owner token. NOT idempotent -- each call mints a new session_id. Never updates or deletes an existing session; the get_* tools read what this writes.
prompt: the user prompt that started the run. model_id: the provider model identifier, e.g. "anthropic.claude-3-5-sonnet-20241022-v2:0". loop_result: the run's token / latency / trace payload. Required keys: input_tokens (int), output_tokens (int), total_tokens (int), latency_ms (float), turns (list). Optional: trace (list), context_blocks (list -- omit if you don't have per-block context data; get_context_timeline needs it).
each turns item: input_tokens (int), output_tokens (int),
latency_ms (float); optional cache_read_input_tokens /
cache_write_input_tokens (int, default 0).
each trace item: tool (str), args (dict), status (str); optional
latency_ms (float, default 0), timestamp (float epoch seconds, default
record time).
each context_blocks item: category (str), label (str), char_count
(int), token_estimate (int); optional turn_n (int or null -- null for a
pre-conversation block), status (str), content (str).
Example loop_result:
{
"input_tokens": 1200, "output_tokens": 340, "total_tokens": 1540,
"latency_ms": 4210.0,
"turns": [
{"input_tokens": 1200, "output_tokens": 340, "latency_ms": 4210.0,
"cache_read_input_tokens": 800, "cache_write_input_tokens": 0}
],
"trace": [
{"tool": "grep_logs", "args": {"pattern": "ERROR"}, "status": "ok",
"latency_ms": 120.0, "timestamp": 1756400000.0}
]
}
| Name | Required | Description | Default |
|---|---|---|---|
| prompt | Yes | ||
| model_id | Yes | ||
| loop_result | Yes |
| Name | Required | Description |
|---|---|---|
| result | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The description explicitly notes non-idempotency, that each call mints a new session_id, that existing sessions are never updated or deleted, and that records are attributed to the connected identity (Google account or owner=None). This complements the annotations, which are structural hints, with concrete side-effect details.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is long, but every section earns its place: purpose and identity semantics are front-loaded, parameter details are systematically organized, and the example makes the nested structure concrete. Its length is proportional to the complexity of the tool.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given a nested parameter object, an existing output schema, and complex optional branches, the description is complete: it covers auth attribution, return value, non-idempotency, required/optional fields, defaults, and a worked example.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 0%, and the description fully compensates: it defines prompt, model_id, and loop_result, and enumerates required and optional nested keys for turns, trace, and context_blocks with a complete JSON example.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
Starts with a specific verb and resource: 'Append ONE agent execution's metrics to this server's store; returns the new session_id.' It clearly identifies this as the write counterpart to the get_* read tools, so an agent can differentiate it without opening the schema.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description states when it is appropriate: 'This is how a remote agent gets its own runs into the server, rather than only being able to query what the server owner recorded locally.' It also clarifies it never updates or deletes and that read access happens through the get_* siblings.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v0.4.0get_agent_traceget_context_timelineget_cost_estimateget_recent_sessionsget_session_metricsget_token_breakdownget_tool_metricsrecord_sessionScored across 8 tools
Each read tool targets a distinct slice of session data—totals, per-turn breakdown, tool-call counts, ordered trace, cost, context timeline, and session discovery—and the descriptions explicitly point to which tool to use. record_session is the only write tool and is clearly separated. No two tools are fungible.
All read tools follow a consistent get_<object>_<metric> snake_case pattern, and the single write tool uses the verb record_session. The naming is predictable and matches each tool's function without mixing conventions.
Eight tools is well-scoped for a session-metrics store: one intake tool, one session-discovery tool, and six focused query/analytics views. Each tool earns its place with no redundancy or bloat.
The server covers the full intended lifecycle: record_session ingests runs, get_recent_sessions discovers session IDs, and the get_* tools provide totals, turn-level detail, tool traces, cost estimates, and context-window composition. Update and delete are deliberately excluded from this append-only metrics store, so there are no obvious dead ends.
MCP server for building and testing AI agents with multi-model experimentation and insights.
MCP observability. Query live traffic, errors, duration, and alerts from your AI agent.
MCP server for your apps' tools and custom tools, plus hosted AI agents and approval-gated workflows
Analytics for MCP servers. Find out which of your tools agents get wrong. MCPulse shows you which tools AI agents retry, which come back empty, and which they never call at all. Two lines inside your own server. It never sees your arguments or your results. getmcpulse.com
