#outputschema
LightNews LightNews
aitoolstack.bsky.social
outputSchema only landed in the 2025-06-18 spec revision; before that tools/list returned name, description, inputSchema, and output shape lived as prose in the description. Does that drift faster than a versioned schema?
November 9, 2026 at 11:18 PM
aperisa.bsky.social
mcp tool tip: declare an outputSchema and the spec says servers MUST return results that conform to it, in structuredContent. put the same json, serialized, in a text block too, for clients that ignore structuredContent.
November 3, 2026 at 7:00 PM
dhanushkumarsivaji.bsky.social
Tip: MCP TypeScript SDK 2.4.0 changes a result shape in a minor. If a tool's outputSchema is z.object().meta({ id }), 2025-11-25 connections drop the { result } wrapper.

Read structuredContent, not .result, and parse it at runtime. A cast won't notice.

#TypeScript #MCP
Code card titled "A minor release that changes a result shape." It covers MCP TypeScript SDK 2.4.0 (client, server and core, Oct 23, 2026). An illustrative TypeScript diff shows reading tool output: the old line reads res.structuredContent.result, the new line parses res.structuredContent with a schema, with the comment "parse, don't cast: a cast would not notice the move". A table of upgrade notes: outputSchema built from z.object(...).meta({ id }) is no longer wrapped in { result } on 2025-11-25 connections; x-mcp-header on a number parameter is rejected per the 2026-07-28 spec, use z.int(); the bundled Ajv uses fast-uri 3.1.8, so an invalid URI scheme like my_tool:output no longer compiles; new Skills over MCP (SEP-2640), where SkillsClient verifies size, digest and SKILL.md frontmatter. Take: read the upgrade notes before bumping a minor, and validate tool output at runtime.
October 25, 2026 at 9:35 AM
worklore.dev
New story: My MCP server's output schemas accepted wrong answers: 7 of 8 tools, found with mcp-drill

MCP lets a tool declare an outputSchema, so a client can check a result before it acts on it.

#mcp #outputschema #jsonschema
https://worklore.dev/s/2026-10-15-my-mcp-server-s-output-schemas-accepted-wrong-answers-7-of-8?utm_source=bluesky&utm_medium=auto
worklore.dev
October 16, 2026 at 1:30 AM
jovidecroock.com
Introducing mcp-tada: typed MCP tool calls for TypeScript.

Snapshot a server's tools/list and every call gets a narrowed tool name, args inferred from inputSchema, and structuredContent typed from outputSchema.

github.com/JoviDeCroock...
GitHub - JoviDeCroock/mcp-tada: Compile-time typed MCP tool calls for TypeScript
Compile-time typed MCP tool calls for TypeScript. Contribute to JoviDeCroock/mcp-tada development by creating an account on GitHub.
github.com
September 10, 2026 at 5:36 PM
happy-homhom.bsky.social
`Tool execution failed` が出るとき、まず見るべきこと
https://papoo.work/doc/6ee1dc196b16b41f
#claudehowto #claudecode #mcp #トラブルシューティング
`Tool execution failed` が出るとき、まず見るべきこと
自分のケースかどうかは、かなりはっきりしています。少なくとも issue 報告では、Claude Desktop / Claude Code 側で MCP の tool call が失敗し、相手のサーバーに届いていない例が出ています。 80105 では、outputSchema を宣言した MCP ツールだけが Tool execution failed になり、サーバー側のログに tools/c
papoo.work
August 24, 2026 at 9:39 AM
happy-homhom.bsky.social
`Tool execution failed` が出るとき、まず見るべきこと
https://papoo.work/doc/6ee1dc196b16b41f
#claudehowto #claudecode #mcp #トラブルシューティング
`Tool execution failed` が出るとき、まず見るべきこと
自分のケースかどうかは、かなりはっきりしています。少なくとも issue 報告では、Claude Desktop / Claude Code 側で MCP の tool call が失敗し、相手のサーバーに届いていない例が出ています。 80105 では、outputSchema を宣言した MCP ツールだけが Tool execution failed になり、サーバー側のログに tools/c
papoo.work
August 23, 2026 at 10:50 PM
happy-homhom.bsky.social
`Tool execution failed` が出るとき、まず見るべきこと
https://papoo.work/doc/6ee1dc196b16b41f
#claudehowto #claudecode #mcp #トラブルシューティング
`Tool execution failed` が出るとき、まず見るべきこと
自分のケースかどうかは、かなりはっきりしています。少なくとも issue 報告では、Claude Desktop / Claude Code 側で MCP の tool call が失敗し、相手のサーバーに届いていない例が出ています。 80105 では、outputSchema を宣言した MCP ツールだけが Tool execution failed になり、サーバー側のログに tools/c
papoo.work
August 21, 2026 at 8:50 AM
horizonshield.bsky.social
3/ Measured from the contract, not behaviour: does a tool's outputSchema declare a boolean or a 2+ enum where that state can live? A spread sample of the 12,429; 18 spoke MCP, 335 tools. Two rulers, opposite errors, so I publish a band.
August 20, 2026 at 9:21 AM
wassim.dev
If you're building #MCP developers in TS and are setting a `outputSchema` to your tools, then you MUST provide a `structuredContent` to your CallToolResult (unless it's an error).

You're Welcome!
September 10, 2025 at 1:54 PM
erisianrite.com
I'm not sure what you're getting at here - performance is from the end user's perspective, not the model's. The model doesn't see anything wrt MCP's outputSchema validation since it's just JSON schema validation in the tool's execution environment on an already-produced response.
September 12, 2025 at 10:07 PM
azu.bsky.social
うーん、これのClientってoutputSchemaのバリデーションやってないのかな "modelcontextprotocol/typescript-sdk: The official Typescript SDK for Model Context Protocol servers and clients" https://github.com/modelcontextprotocol/typescript-sdk
July 3, 2025 at 2:27 PM
azu.bsky.social
見てる: "`2025-06-18` Discrepancy: `structuredContent` does not support full `outputSchema` JSON Schema · Issue #834 · modelcontextprotocol/modelcontextprotocol" https://github.com/modelcontextprotocol/modelcontextprotocol/issues/834
June 29, 2025 at 3:38 PM
community.openai.com.web.brid.gy
Output schema in mcp is not typed in agent builder [ {{input.results}} type issue ]
Added an mcp server in agent builder, the problem is the output schema from the tool call is not respected - I have defined the output schema and return the results as expected for reference outputSchema: { name: z.string().describe("The name of the speaker"), jobTitle: z.string().describe("The job title of the speaker"), company: z.string().describe("The company of the speaker"), }, and this is what is returned from the mcp server { content: [ { type: "text", text: JSON.stringify(HARDCODED_SPEAKER, null, 2) } ], structuredContent: HARDCODED_SPEAKER, toolResult: HARDCODED_SPEAKER }; hardcoded_speaker is an object - Everything works fine if I add an additional agent and extract the values of hardcoded speaker — MCP node –> Agent The core issue is {{input.results}} is not typed
community.openai.com
October 15, 2025 at 12:13 PM
azukiazusa.dev.web.brid.gy
MCP の Structured tool output を試してみる
MCP の 2025-06-18 バージョンでは Structured tool output がサポートされました。ツールの定義で `outputSchema` を出力のスキーマを定義し、`structuredContent` フィールドに構造化された出力を返すことができます。この記事では MCP の TypeScript SDK を使用して Structured tool output を試してみます。
azukiazusa.dev
July 10, 2025 at 8:50 AM
azukiazusa.dev.web.brid.gy
MCP の Structured tool output を試してみる
MCP の 2025-06-18 バージョンでは Structured tool output がサポートされました。ツールの定義で `outputSchema` を出力のスキーマを定義し、`structuredContent` フィールドに構造化された出力を返すことができます。この記事では MCP の TypeScript SDK を使用して Structured tool output を試してみます。
azukiazusa.dev
July 8, 2025 at 8:50 AM
azukiazusa.dev.web.brid.gy
MCP の Structured tool output を試してみる
MCP の 2025-06-18 バージョンでは Structured tool output がサポートされました。ツールの定義で `outputSchema` を出力のスキーマを定義し、`structuredContent` フィールドに構造化された出力を返すことができます。この記事では MCP の TypeScript SDK を使用して Structured tool output を試してみます。
azukiazusa.dev
July 1, 2025 at 8:49 AM
azukiazusa.dev.web.brid.gy
MCP の Structured tool output を試してみる
MCP の 2025-06-18 バージョンでは Structured tool output がサポートされました。ツールの定義で `outputSchema` を出力のスキーマを定義し、`structuredContent` フィールドに構造化された出力を返すことができます。この記事では MCP の TypeScript SDK を使用して Structured tool output を試してみます。
azukiazusa.dev
June 30, 2025 at 8:52 AM
azukiazusa.dev.web.brid.gy
MCP の Structured tool output を試してみる
MCP の 2025-06-18 バージョンでは Structured tool output がサポートされました。ツールの定義で `outputSchema` を出力のスキーマを定義し、`structuredContent` フィールドに構造化された出力を返すことができます。この記事では MCP の TypeScript SDK を使用して Structured tool output を試してみます。
azukiazusa.dev
June 29, 2025 at 8:51 AM
gunad.bsky.social
just submitted an MCP server as a ChatGPT plugin.

what they require:
→ Stripe identity verify (said 90 days, took 3)
→ OAuth 2.0 + PKCE
→ 3 annotations per tool
→ outputSchema on every tool
→ 5 positive + 3 negative test cases

already had MCP for Claude 80% was done.
July 13, 2026 at 3:57 AM
zuplo.com
The schema is defined by the outputSchema property on your tool, which clients can use to validate your outputs. Clients are going to start keeping you honest! 🤯
June 20, 2025 at 5:56 PM
pipeworx.io
Connect any agent to gateway.pipeworx.io/mcp and it sees:

• 1,007 tools / 295 packs (SEC, FDA, FRED, Census, EPA, ATTOM, ClinicalTrials…)
• Every tool has inputSchema examples + outputSchema
• 15 Resource templates (citable URIs)
• 12 Prompt playbooks
• 5 meta-tools
gateway.pipeworx.io
May 8, 2026 at 8:51 PM
chatforest.bsky.social
New guide: MCP Structured Output Deep Dive — how outputSchema and structuredContent give MCP tools real return types. Covers schema design, dual responses, validation, tool chaining, and migration. https://chatforest.com/guides/mcp-structured-output-deep-dive/
MCP Structured Output Deep Dive
How outputSchema and structuredContent turn MCP tools into typed functions — schema design, validation, tool chaining, and migration.
chatforest.com
March 27, 2026 at 10:49 PM
chatforest.bsky.social
Designing MCP tools that agents actually use well — new guide covers outputSchema, structured content, agent-aware response patterns (summary-first, error recovery, confidence scoring), composability, and naming conventions. https://chatforest.com/guides/mcp-tool-design-patterns/
MCP Tool Design Patterns: Building Agent-Friendly, Composable Tools
How to design MCP tools that agents can discover, understand, compose, and recover from.
chatforest.com
March 27, 2026 at 9:58 PM
karashiiro.moe
oh yeah also there's the GitHub MCP server which returns JSON in a text block (and not also in structuredContent) so it's not allowed to have an outputSchema, which I do attempt to handle but I think that should be a crime in some jurisdictions
December 20, 2025 at 6:26 PM