agentleFS
truefoundrytruefoundry/skills/skills/mcp-servers/SKILL.md

truefoundry-mcp-servers

Skill14 starsChanged 6 months agoMITView on GitHub

Manages TrueFoundry MCP Gateway server registry entries. Covers listing, creating, updating, tool selection, access control, and client attachment for remote, official remote, virtual, hosted STDIO, and OpenAPI-backed MCP servers. Delete operations are manual dashboard-only.

The file

Show its rules and text

We split this skill into the rules it gives an agent

40 rules. About 0 are common, 26 are distinctive, and 14 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈26

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 14

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use this skill when the user wants to:

    line 15
  2. Distinctive:

    Create virtual MCP servers from tools across multiple source servers.

    line 19
  3. Very specific:

    Create hosted STDIO MCP servers from mcpServers JSON.

    line 20
  4. Distinctive:

    Import OpenAPI specs as MCP servers.

    line 21
  5. Distinctive:

    Update auth, access, metadata, endpoint/spec, or selected tools.

    line 23
  6. Distinctive:

    Add MCP servers to clients such as Cursor, VS Code, Claude Code, Python, TypeScript, Windsurf, or Codex.

    line 24
  7. Distinctive:

    Do not delete MCP servers or remove collaborators from the agent.

    line 26
  8. Very specific:

    Treat the live dashboard Apply using YAML preview as the source of truth when it differs from examples.

    line 41
  9. Very specific:

    Verify tfy login is complete.

    line 45
  10. Very specific:

    Verify the target tenant from TFY_BASE_URL/TFY_HOST.

    line 46
  11. Very specific:

    Use mcp-server-manager as the default role unless the user requests another role.

    line 48
  12. Very specific:

    Use collaborator subjects as user:<email>, team:<slug>, serviceaccount:<name>, or virtualaccount:<name>.

    line 49
  13. Very specific:

    Use tfy-secret://... references for shared credentials whenever possible.

    line 50
  14. Distinctive:

    Never print, paste, or reveal real credentials unless the user explicitly asks and accepts exposure.

    line 51
  15. Distinctive:

    Do not call DELETE APIs, destructive CLI commands, or collaborator-removal flows.

    line 52
  16. Very specific:

    Before every final tfy apply, show:

    line 54
  17. Distinctive:

    Prefer the dashboard when the user is working in the UI.

    line 65
  18. Distinctive:

    Use this for an existing HTTP MCP endpoint.

    line 81
  19. Distinctive:

    if auth is enabled: API key/header, OAuth2, or token passthrough

    line 90
  20. Very specific:

    No auth: omit auth_data.

    line 94
  21. Distinctive:

    Use references/mcp-yaml-variations.md for concrete YAML shapes.

    line 99
  22. Distinctive:

    Use this for catalog servers such as GitHub, Linear, Sentry, Atlassian, Figma, Slack, Notion, Stripe, HuggingFace, or similar.

    line 103
  23. Distinctive:

    Use this to expose selected tools from multiple existing MCP servers through one virtual MCP server.

    line 117
  24. Distinctive:

    Ask which tools to expose from each source.

    line 133
  25. Very specific:

    Use servers[].enabled_tools for selected subsets.

    line 134
  26. Very specific:

    Add collaborators with mcp-server-manager unless requested otherwise.

    line 136
  27. Distinctive:

    Use references/mcp-yaml-variations.md for virtual server YAML examples.

    line 140
  28. Distinctive:

    Use this when the user has a local-style MCP config and wants the gateway to host/expose it.

    line 144
  29. Distinctive:

    Ask for the STDIO JSON or file.

    line 150
  30. Distinctive:

    Use the gateway import flow or YAML preview as the source of truth for the manifest generated after import.

    line 152
  31. Distinctive:

    Use references/mcp-yaml-variations.md for the input shape.

    line 155
  32. Distinctive:

    Use this to expose OpenAPI operations as MCP tools.

    line 159
  33. Very specific:

    Do not invent field names for normal-server tool filtering.

    line 186
  34. Distinctive:

    Use secret references for shared credentials.

    line 197
  35. Distinctive:

    Use references/mcp-yaml-variations.md for auth examples.

    line 201
  36. Distinctive:

    When changing access, show the before/after collaborator list and preserve unrelated server config.

    line 213
  37. Very specific:

    Use the dashboard Add to Client modal or the generated snippet.

    line 219
  38. Distinctive:

    Always run the dry-run first:

    line 223
  39. Distinctive:

    Run the final apply only after explicit user confirmation:

    line 229
  40. Very specific:

    Do not call DELETE APIs or tfy delete flows.

    line 237

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md
<objective>

MCP Servers

Manage MCP servers in TrueFoundry AI Gateway.

Use this skill when the user wants to:

  • List MCP servers attached to a gateway.
  • Connect remote or official remote MCP servers.
  • Create virtual MCP servers from tools across multiple source servers.
  • Create hosted STDIO MCP servers from mcpServers JSON.
  • Import OpenAPI specs as MCP servers.
  • Enable or disable tools.
  • Update auth, access, metadata, endpoint/spec, or selected tools.
  • Add MCP servers to clients such as Cursor, VS Code, Claude Code, Python, TypeScript, Windsurf, or Codex.

Do not delete MCP servers or remove collaborators from the agent. Route those requests to the TrueFoundry dashboard.

</objective><instructions>

Reference Map

Load only the reference needed for the user's task:

TaskReference
Browser/UI operation, click paths, form fields, add-to-client modalreferences/mcp-gateway-ui-flows.md
YAML examples for auth modes, server types, virtual tool selection, OpenAPI, hosted STDIOreferences/mcp-yaml-variations.md

Treat the live dashboard Apply using YAML preview as the source of truth when it differs from examples.

Safety Rules

  • Verify tfy login is complete. If not, use truefoundry-onboard.
  • Verify the target tenant from TFY_BASE_URL/TFY_HOST.
  • Confirm the workspace or registry scope with the user before applying changes.
  • Use mcp-server-manager as the default role unless the user requests another role.
  • Use collaborator subjects as user:<email>, team:<slug>, serviceaccount:<name>, or virtualaccount:<name>.
  • Use tfy-secret://... references for shared credentials whenever possible.
  • Never print, paste, or reveal real credentials unless the user explicitly asks and accepts exposure.
  • Do not call DELETE APIs, destructive CLI commands, or collaborator-removal flows.

Before every final tfy apply, show:

  • Plain-English summary of the change.
  • Target tenant and workspace/scope.
  • Full YAML or diff.
  • Exact command to be run.

Then ask for explicit confirmation. Run only dry-runs before confirmation.

List MCP Servers

Prefer the dashboard when the user is working in the UI. For API fallback:

TFY_API_SH=~/.claude/skills/truefoundry-mcp-servers/scripts/tfy-api.sh
$TFY_API_SH GET /api/svc/v1/mcp-servers

Present servers as:

MCP Servers
| Name | Type | Auth | ID | URL / Source |

Connect Remote MCP Server

Use this for an existing HTTP MCP endpoint.

Collect:

  • name
  • description
  • url
  • collaborators and roles
  • auth enabled or disabled
  • if auth is enabled: API key/header, OAuth2, or token passthrough

Auth rules:

  • No auth: omit auth_data.
  • API key/header auth: support shared credentials and individual credentials; require at least one header.
  • OAuth2: support authorization code and client credentials; include optional scopes, PKCE/code challenge methods, Dynamic Client Registration URL, and additional token params only when supplied or shown by the UI preview.
  • Token passthrough: use the caller's existing TrueFoundry token; no extra fields are needed.

Use references/mcp-yaml-variations.md for concrete YAML shapes.

Connect Official Remote MCP Server

Use this for catalog servers such as GitHub, Linear, Sentry, Atlassian, Figma, Slack, Notion, Stripe, HuggingFace, or similar.

Flow:

  1. Identify the official server in the gateway catalog.
  2. Confirm the catalog auth mode.
  3. For OAuth2, tell the user they may need to complete a browser authentication flow after creation.
  4. For header auth, collect header names and use secret references or per-user placeholders.
  5. Confirm collaborators.
  6. Generate YAML from the UI preview or the YAML reference.
  7. Dry-run, show diff, then apply only after confirmation.

Create Virtual MCP Server

Use this to expose selected tools from multiple existing MCP servers through one virtual MCP server.

Collect:

  • virtual MCP name
  • description
  • source MCP servers
  • tools to expose from each source server
  • collaborators and roles
  • optional source-server auth header override if the gateway requires it

Flow:

  1. List registered MCP servers.
  2. Select source MCP servers.
  3. Enumerate available tools for each source server.
  4. Ask which tools to expose from each source.
  5. Use servers[].enabled_tools for selected subsets.
  6. Omit enabled_tools for a source only when exposing every tool from that source.
  7. Add collaborators with mcp-server-manager unless requested otherwise.
  8. Show virtual server name, source servers, selected tools, and access.
  9. Show YAML, dry-run, and apply only after confirmation.

Use references/mcp-yaml-variations.md for virtual server YAML examples.

Create Hosted STDIO MCP Server

Use this when the user has a local-style MCP config and wants the gateway to host/expose it.

Collect JSON containing an mcpServers object with server command, args, and optional env.

Flow:

  1. Ask for the STDIO JSON or file.
  2. Replace raw environment secrets with secret references where supported, or tell the user which values must be moved to secrets.
  3. Use the gateway import flow or YAML preview as the source of truth for the manifest generated after import.
  4. Show summary, YAML/diff, dry-run, and confirmation-gated apply.

Use references/mcp-yaml-variations.md for the input shape.

Import From OpenAPI Spec

Use this to expose OpenAPI operations as MCP tools.

Collect:

  • name
  • description
  • OpenAPI spec source: trusted remote URL or pasted inline spec
  • selected tools/operations
  • collaborators
  • optional auth using the same modes as remote MCP servers

For remote specs, confirm the URL is trusted before use. Prefer pasted/inline specs for sensitive private APIs.

Enable Or Disable Tools

Tool filtering is supported for normal remote/OpenAPI servers and virtual MCP servers.

For normal remote/OpenAPI servers:

  1. Fetch the current MCP server config by ID.
  2. Inspect the API response or UI YAML preview for the existing tool-selection shape.
  3. Fetch or confirm available tool names.
  4. Modify only the requested tool-selection fields.
  5. Preserve name, URL/spec, auth, collaborators, TLS, tags, and unrelated settings.
  6. Show the before/after tool list or YAML diff.
  7. Dry-run and apply only after confirmation.

Do not invent field names for normal-server tool filtering. If the API response does not expose the shape, use the dashboard edit screen or ask the user for the Apply using YAML output.

For virtual servers, use servers[].enabled_tools; omit it only when exposing every tool from that source.

Update Auth

Use update-by-manifest:

  1. Fetch the current MCP server config.
  2. Preserve name, URL/spec, selected tools, collaborators, TLS, tags, and unrelated settings.
  3. Replace only the requested auth_data block.
  4. Use secret references for shared credentials.
  5. Show old auth mode and new auth mode.
  6. Show YAML/diff, dry-run, and apply only after confirmation.

Use references/mcp-yaml-variations.md for auth examples.

Access Control

Adding collaborators is allowed with confirmation. Removing collaborators is dashboard-only.

To list access:

$TFY_API_SH GET '/api/svc/v1/collaborators?resourceType=mcp-server&resourceId=RESOURCE_ID'

When changing access, show the before/after collaborator list and preserve unrelated server config.

Add To Client

The gateway UI can generate setup snippets for Cursor, VS Code, Claude Code, Python, TypeScript, Windsurf, and Codex.

Use the dashboard Add to Client modal or the generated snippet. Do not click, reveal, paste, or print API keys unless the user explicitly asks and understands the exposure.

Apply

Always run the dry-run first:

tfy apply -f mcp-server.yaml --dry-run --show-diff

Run the final apply only after explicit user confirmation:

tfy apply -f mcp-server.yaml

Delete

Do not call DELETE APIs or tfy delete flows. Tell the user:

To delete this MCP server, open the TrueFoundry dashboard, go to AI Gateway -> MCP Servers, select the server, and delete it from the UI.
</instructions>

<success_criteria>

  • The user can list MCP servers.
  • The user can create remote, official remote, virtual, OpenAPI, and hosted STDIO-backed MCP servers.
  • Auth-disabled, API key, OAuth2, and token-passthrough configurations match the live gateway YAML shape.
  • Tool enable/disable updates preserve unrelated server config.
  • Auth and access changes use reviewed full-manifest updates.
  • Client attachment guidance covers Cursor, VS Code, Claude Code, Python, TypeScript, Windsurf, and Codex.
  • Shared credentials use tfy-secret://... references or explicit placeholders.
  • Final tfy apply is gated by summary, YAML/diff, exact command, and explicit confirmation.
  • Deletes and collaborator removals are dashboard-only.

</success_criteria>

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.