everything server: instructions unconditionally reference sampling/elicitation tools that only exist when the client declares those capabilities
Nobody has claimed this yet.
Assessment
- Difficulty
- 2/5
- Estimated time
- 1-3 hours
- Newbie friendliness
- 75/100
- Issue type
- Bug
- Clarity
- Mostly clear
- Activity status
- Active
- Tech stack
- typescript
- Domain
- backend
Research direction
Start in server/index.ts, focusing on registerConditionalTools() inside oninitialized and the static instructions string. Run the provided npx server reproduction with empty and sampling/elicitation capabilities, then verify that the instructions consistently describe whether trigger-sampling-request and trigger-elicitation-request are available.
Written by the indexing model from the issue text.
Description
Factual issue
The @modelcontextprotocol/server-everything v2026.8.31 instructions string unconditionally tells integrating agents to use trigger-sampling-request and trigger-elicitation-request:
## Constraints & Limitations
- Sampling requests (`trigger-sampling-request`) require client sampling capability
- Elicitation requests (`trigger-elicitation-request`) require client elicitation capability
These tools are registered through registerConditionalTools() inside oninitialized, gated on the client's declared capabilities (server/index.ts). For a client that declares no sampling/elicitation capability — e.g. the majority of clients today, and verified by a capability sweep across 12 popular MCP servers — the tools never appear in tools/list, and the instructions reference tools that do not exist for that session:
- initialize with
capabilities: {}→tools/listreturns 13 tools, none of themtrigger-sampling-requestortrigger-elicitation-request - initialize with
capabilities: {sampling:{}, elicitation:{}}→ 16 tools, both present
Suggested fixes (pick any)
- Make the instructions conditional-aware — e.g. "when your client declares the sampling capability,
trigger-sampling-requestis available" — so the guidance matches what an agent can actually call. - Compose the instructions per session — the instructions string is static, but it could be built from the registered-tools state after
oninitialized, or the conditionality could be stated in one sentence. - At minimum, add the phrase "if your client declares the corresponding capability" to the two bullets.
Why this matters beyond this server
The instructions field exists to orient agents; an agent that follows these instructions on a non-capability client will attempt to call a tool that is not in its catalog — a coherence class that deterministic batteries (this was found with mcp-eval's catalog-vs-instructions checks) can expose mechanically.
Repro
npx -y @modelcontextprotocol/server-everything
# initialize with capabilities: {} ; then tools/list -> 13 tools, no trigger-sampling-request
# initialize with capabilities: {sampling:{},elicitation:{}} ; wait ~1.5s after notifications/initialized ; tools/list -> 16 tools, both present
Environment: server-everything v2026.8.31 (npm), server self-reports 2.0.0; verified against the published dist (registerConditionalTools in dist/server/index.js).
- Dominant language
- TypeScript
- Stars
- 90.5k
- Forks
- 11.7k
- Avg merge
- 5d 19h
- Merged PRs (30d)
- 6
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from modelcontextprotocol/servers
-
Difficulty 2/5 1-3 hours Newbie friendliness 72/100
modelcontextprotocol/servers#4830 ·
-
Difficulty 2/5 1-3 hours Newbie friendliness 68/100
modelcontextprotocol/servers#4829 ·
-
Difficulty 2/5 1-3 hours Newbie friendliness 70/100
modelcontextprotocol/servers#4812 ·
-
Difficulty 2/5 1-3 hours Newbie friendliness 86/100
modelcontextprotocol/servers#4804 · 1 comment ·
-
README gate confirmation command can be used by any commenter, bypassing the "readme: pending" gate Open
Difficulty 2/5 1-3 hours Newbie friendliness 82/100
modelcontextprotocol/servers#4796 ·
All issues in modelcontextprotocol/servers
Similar issues
-
enhancement
Difficulty 2/5 1-3 hours Newbie friendliness 70/100
dennys-bd/agent-hive#184 ·
-
Add: hunch Open
Difficulty 2/5 1-3 hours Newbie friendliness 74/100
AbdelStark/awesome-typesafe#104 ·
-
ai-observability bug team/ai-observability
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
-
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
-
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
vicharanashala/fln#563 ·