mirror of
https://github.com/paperclipai/paperclip.git
synced 2026-11-20 00:53:46 +01:00
## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - Agents use connected apps through MCP. Paperclip already hosts MCP Apps views and speaks stateless MCP 2026-07-28. > - The MCP Skills extension (SEP-2640, Final) lets an app publish skills (instructions and files) next to its tools. Hugging Face already serves 25. > - Paperclip did not read them, so agents missed each app's own workflow guidance. > - Skills are a prompt-injection surface. The spec makes manifest binding, verification, origin tagging and approval revocation host requirements. > - This pull request makes Paperclip a Skills host: discover, review, verify, and deliver app skills to the agents that use the app. > - The benefit is that connected apps can teach agents their workflows, under board control. ## Linked Issues or Issue Description **Problem** Connected MCP servers can publish skills with the `io.modelcontextprotocol/skills` extension. Paperclip ignored them. A board could not see, review or approve them, and agents never received them. **Proposed solution** Implement the host side of SEP-2640. List skills during catalog refresh, record them per connection and company, bind board approval to the full file manifest, and load approved skills into agent runs with verified, lazily fetched content. **Why it matters** Apps ship workflow guidance with their tools. Without the extension, each team must copy that guidance into company skills by hand, and it goes stale. Refs #21119, Refs #21140 (the MCP Apps and stateless MCP client work this builds on). ## What Changed - **Client capability.** Every stateless request declares `io.modelcontextprotocol/skills`. Legacy handshake sessions do not. The capabilities that `server/discover` advertises are remembered beside the protocol era. - **Discovery.** Catalog refresh calls `skills/list` (paged) only when the server declared the extension. A listing inside its `ttlMs` is reused. A skills failure never fails the tool refresh. Nothing runs while a run starts. - **Schema.** New `tool_mcp_skills` (company, connection, skill URI; frontmatter, sorted manifest, manifest digest, review state) and `tool_mcp_skill_files` (verified bytes keyed by connection, URI and digest). Both cascade with the connection. - **Governance.** New skills wait for review. Approval binds to the order-free digest of every `{uri, digest, size}`; a request for any other digest gets 409. A changed manifest or frontmatter revokes approval and logs `tool_mcp_skill.approval_revoked`. Dynamic, oversized and invalid skills are declined with a reason. Approve and decline use the company skill policy (`skills.install`, `external_package`, `mcp-connection:<id>`). - **Run time.** Approved skills of connections the agent can use load in parallel with other skill preparation, inside a 3 s budget. `SKILL.md` is read with the agent's grant and checked for size, SHA-256 and frontmatter. A failure calls `skills/get` and takes the skill out of use. Each skill becomes a per-run bundle on the connector skill path, named in its app's namespace, tagged with its origin. Paperclip writes the frontmatter: `allowed-tools` only if the board approved it. - **Supporting files.** `GET /api/agents/me/mcp-skills/:skillId/files?path=` serves manifest-listed files to an agent in an active run, through the skill's own connection. The sandbox bridge allows the route. - **Board API and UI.** `GET/POST /api/companies/:companyId/mcp-skills…` (list, read instructions, approve, decline). The Skills page shows "Skills from connected apps" with app, state, size, requested tools and plain-language notes. It renders nothing when no app publishes skills. - **Docs.** `doc/connections/MCP-SKILLS.md`, linked from `doc/connections/README.md`. - **Shared frontmatter parser.** It now decodes YAML single-quoted scalars. Before, 5 of Hugging Face's 25 skills kept their quotes and would fail the frontmatter check; now all 25 match. The generated runner copy is regenerated. File size notes: `server/src/services/tool-gateway.ts` grows by about 200 lines. The run-time reads must use its grant resolution and `callAssignedConnectionProtocol`, like MCP Apps view reads. Splitting the gateway is a separate refactor. `CompanySkills.tsx` and `CompanySkills.production.tsx` gain 2–4 lines each; the section lives in its own component. ## Verification - `server/src/__tests__/mcp-skills.test.ts` (15 tests) uses a fake stateless skills server and embedded Postgres. It covers: discovery only after the server declares the extension; the capability in `_meta`; `ttlMs` reuse; lazy fetch (no reads at listing or approval; `SKILL.md` at load; supporting files on read); a stale-digest approval refused; a reorder that keeps approval; a changed or added file that revokes it; dynamic skills declined; a same-size tampered `SKILL.md` rejected, refreshed with `skills/get`, and taken out of use; mismatched frontmatter rejected; reads outside the manifest refused; the same URI and name from two servers kept apart; company scoping; a slow app skipped within the budget and cached later; HTTP authorization (board only, agent needs an active run, run header must match) and activity logging. - Red/green: with the digest check disabled, the tamper test fails; with the manifest comparison disabled, the revocation test fails. Both pass with the code. - `server/src/__tests__/mcp-protocol-versions.test.ts`, `mcp-apps.test.ts`, `connection-instructions.test.ts`, `tool-gateway-discovery-http.test.ts`: 47 pass (two capability assertions updated for the new extension). - `packages/shared/src/frontmatter.test.ts` adds a single-quoted scalar case; the runner contract test passes (23 tests). - `ui/src/components/skills/ConnectedAppSkills.test.tsx` (3) and `ui/src/pages/CompanySkills.test.tsx` (44) pass. `packages/adapter-utils/src/sandbox-callback-bridge.test.ts` (89) passes. - `pnpm --filter @paperclipai/server typecheck` passes. UI `tsc` passes except the known missing generated Storybook catalog files. - `pnpm verify` returned exit 2 (doc change is a global CI input); required CI supplies broad coverage. - Live check against `https://huggingface.co/mcp` without credentials, through Paperclip's own services and the real egress guard: 25 skills discovered as pending; three approved; cold load 135 ms, cached load 13 ms; a supporting file read through the manifest; the bundle tagged with its origin. - Storybook: `Agents and reusable capabilities/Skills and Skill Studio/Connected app skills` (needs review, all states, tool request, name collision, reading instructions). ## Risks - **Migration:** two new tables only. No changes to existing tables. Safe with the previous release running. - **Run start:** companies without an approved app skill pay one indexed read. With approved skills the wait is capped at 3 s; late or failing skills are left out with a `skills.mcp_unavailable` run warning and never fail the run. - **Agent harness enforcement:** Paperclip cannot gate each code-execution tool call an agent makes while following an app's skill. Approval is explicit and per skill, the approval note says agents may run commands the skill describes, and `allowed-tools` is never granted without a separate tick. - **Shared-home adapters** receive app skills like company skills. A revoked skill leaves on the next skill sync. - Not in this PR: per-connection opt-in for dynamic skills; serving Paperclip's own skills from its public MCP endpoint; `resources/directory/read`; app skills on an agent's skills tab. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, via Claude Code with tool use (shell, file edits, local test runs, live HTTP checks). Extended reasoning enabled. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [ ] All Paperclip CI gates are green - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
231 lines
12 KiB
TypeScript
231 lines
12 KiB
TypeScript
import { createHash } from "node:crypto";
|
|
import type { Db } from "@paperclipai/db";
|
|
import { deduplicateToolDescriptors, ToolGatewayHttpError, type ToolGatewayDescriptor, type ToolGatewayService } from "../tool-gateway.js";
|
|
|
|
type WorkMode = "standard" | "planning" | "ask" | "skill_test";
|
|
type ToolDefinition = Record<string, unknown>;
|
|
|
|
const SEARCH_TOOL = "paperclip_search_assigned_tools";
|
|
const CALL_TOOL = "paperclip_call_assigned_tool";
|
|
// A single valid runner schema can be 512 KiB. Leave room for its description
|
|
// while keeping the complete result below the 768 KiB provider-result bound.
|
|
const SEARCH_PAGE_BYTES = 640 * 1024;
|
|
const SCHEMA_CHUNK_CHARACTERS = 64 * 1024;
|
|
const ON_DEMAND_TOOLS: ToolDefinition[] = [
|
|
{
|
|
name: SEARCH_TOOL,
|
|
description: "Search your assigned app tools by name or description and read their input schemas. Pass {\"query\":\"search terms\"}, or an empty query to browse. Use limit from 1 to 20 (default 5). Omit unused optional fields instead of sending null. To fetch the next page, pass the returned nextOffset as offset. A large definition returns inputSchemaRef: set schemaTool to that name with an empty query, then pass the returned nextSchemaOffset as schemaOffset until null. Concatenate schemaJson chunks and parse JSON. Use paperclip_call_assigned_tool with a returned name and arguments matching its inputSchema.",
|
|
inputSchema: {
|
|
type: "object", additionalProperties: false,
|
|
properties: {
|
|
query: { type: "string", maxLength: 200 },
|
|
offset: { type: "integer", minimum: 0 },
|
|
limit: { type: "integer", minimum: 1, maximum: 20, default: 5, description: "Maximum results per page (1 to 20; default 5)." },
|
|
schemaTool: { type: "string", description: "Exact inputSchemaRef from a search result. Omit or leave blank for a normal tool search." },
|
|
schemaOffset: { type: "integer", minimum: 0, description: "Schema chunk offset. A positive offset requires an exact inputSchemaRef in schemaTool." },
|
|
},
|
|
required: ["query"],
|
|
},
|
|
},
|
|
{
|
|
name: CALL_TOOL,
|
|
description: "Call an assigned app tool discovered with paperclip_search_assigned_tools. Use its exact returned name and arguments matching its inputSchema. The same permissions, approvals, and audit rules apply as for direct app tools.",
|
|
inputSchema: {
|
|
type: "object", additionalProperties: false,
|
|
properties: { name: { type: "string" }, arguments: { type: "object", additionalProperties: true } },
|
|
required: ["name", "arguments"],
|
|
},
|
|
},
|
|
];
|
|
|
|
function object(value: unknown): Record<string, unknown> {
|
|
if (!value || typeof value !== "object" || Array.isArray(value)) {
|
|
throw new Error("assigned_mcp_tool_invalid_arguments");
|
|
}
|
|
return value as Record<string, unknown>;
|
|
}
|
|
|
|
function definition(name: string, tool: ToolGatewayDescriptor): ToolDefinition {
|
|
return { name, description: `${tool.displayName}: ${tool.description}`, inputSchema: structuredClone(tool.parametersSchema) };
|
|
}
|
|
|
|
// Execution must use the app's configured gateway, including deployment
|
|
// restrictions, OAuth refresh, and approval delivery. Never fall back to an
|
|
// isolated gateway whose defaults differ from the running instance.
|
|
const assignedMcpGateways = new WeakMap<Db, ToolGatewayService>();
|
|
|
|
export function registerAssignedMcpGateway(db: Db, gateway: ToolGatewayService): void {
|
|
assignedMcpGateways.set(db, gateway);
|
|
}
|
|
|
|
export function getAssignedMcpGateway(db: Db): ToolGatewayService {
|
|
const gateway = assignedMcpGateways.get(db);
|
|
if (!gateway) throw new Error("assigned_mcp_gateway_unavailable");
|
|
return gateway;
|
|
}
|
|
|
|
/** No configured gateway means no authorized connection tools or instructions. */
|
|
export async function resolveAssignedConnectionInstructionsForRun(db: Db, binding: { companyId: string; agentId: string; runId: string }) {
|
|
const gateway = assignedMcpGateways.get(db);
|
|
return gateway ? gateway.resolveConnectionInstructionsForRun(binding) : null;
|
|
}
|
|
|
|
/** MCP Skills: approved skills of the agent's usable connections; none without a configured gateway. */
|
|
export async function resolveMcpConnectionSkillsForRun(db: Db, binding: { companyId: string; agentId: string; runId: string }) {
|
|
const gateway = assignedMcpGateways.get(db);
|
|
return gateway ? gateway.resolveConnectionSkillsForRun(binding) : null;
|
|
}
|
|
|
|
function assignedToolName(name: string): string {
|
|
const sanitize = (value: string) => value.replace(/[^a-zA-Z0-9_]+/g, "_").replace(/^_+|_+$/g, "");
|
|
const separator = name.lastIndexOf(":");
|
|
const readable = separator >= 0
|
|
? `${sanitize(name.slice(0, separator)).slice(0, 20) || "tool"}_${sanitize(name.slice(separator + 1)).slice(-26) || "action"}`
|
|
: sanitize(name).slice(0, 47) || "tool";
|
|
const digest = createHash("sha256").update(name).digest("hex").slice(0, 12);
|
|
return `app_${readable}_${digest}`;
|
|
}
|
|
|
|
/** Server-owned MCP bindings projected into the runner's existing tool channel. */
|
|
export async function createAssignedMcpTools(input: {
|
|
gateway: ToolGatewayService;
|
|
gatewayPublicId: string;
|
|
bearerToken: string;
|
|
workMode?: WorkMode;
|
|
/** Bound provider grammar complexity independently of protocol byte capacity. */
|
|
delivery?: "on_demand";
|
|
}) {
|
|
const listed = await input.gateway.listToolsForNamedGateway({
|
|
gatewayPublicId: input.gatewayPublicId,
|
|
bearerToken: input.bearerToken,
|
|
});
|
|
const tools = new Map<string, ToolGatewayDescriptor>();
|
|
for (const descriptor of deduplicateToolDescriptors(listed)) {
|
|
const name = assignedToolName(descriptor.name);
|
|
const previous = tools.get(name);
|
|
if (previous) {
|
|
throw new ToolGatewayHttpError(409,
|
|
`Assigned MCP tool name collision for ${JSON.stringify([previous.name, descriptor.name])}.`,
|
|
"assigned_mcp_tool_name_collision");
|
|
}
|
|
tools.set(name, descriptor);
|
|
}
|
|
const permits = (tool: ToolGatewayDescriptor, mode: WorkMode = input.workMode ?? "standard") => mode === "standard" || tool.risk === "read";
|
|
// OpenCode's HTTP bridge returns at most 64K JSON characters. Leave room
|
|
// for the envelope and for JSON escaping when returning a schema chunk.
|
|
// Other harnesses retain the larger PRP page and chunk budgets.
|
|
const searchPageBytes = input.delivery === "on_demand" ? 32 * 1024 : SEARCH_PAGE_BYTES;
|
|
const schemaChunkCharacters = input.delivery === "on_demand" ? 16 * 1024 : SCHEMA_CHUNK_CHARACTERS;
|
|
|
|
async function search(argumentsValue: unknown, currentWorkMode?: WorkMode) {
|
|
const args = object(argumentsValue);
|
|
const offset = args.offset ?? 0;
|
|
const limit = args.limit ?? 5;
|
|
const schemaOffset = args.schemaOffset ?? 0;
|
|
const hasSchemaTool = typeof args.schemaTool === "string" && args.schemaTool.trim().length > 0;
|
|
if (typeof args.query !== "string" || args.query.length > 200 ||
|
|
typeof offset !== "number" || !Number.isSafeInteger(offset) || offset < 0 ||
|
|
typeof limit !== "number" || !Number.isSafeInteger(limit) || limit < 1 || limit > 20 ||
|
|
(args.schemaTool !== undefined && typeof args.schemaTool !== "string") ||
|
|
typeof schemaOffset !== "number" || !Number.isSafeInteger(schemaOffset) || schemaOffset < 0) {
|
|
throw new Error("assigned_mcp_tool_invalid_arguments");
|
|
}
|
|
if (schemaOffset > 0 && !hasSchemaTool) {
|
|
throw new Error("assigned_mcp_tool_invalid_arguments: A positive schemaOffset requires an exact inputSchemaRef in schemaTool. Omit schemaTool and schemaOffset to search assigned tools.");
|
|
}
|
|
// Intersect fresh grants with this session's pinned catalog. Revocations
|
|
// take effect immediately; newly assigned tools require a new session.
|
|
const current = new Map(deduplicateToolDescriptors(await input.gateway.listToolsForNamedGateway({
|
|
gatewayPublicId: input.gatewayPublicId, bearerToken: input.bearerToken,
|
|
})).map(tool => [tool.name, tool]));
|
|
const terms = args.query.toLowerCase().trim().split(/\s+/).filter(Boolean);
|
|
const authorized = [...tools].filter(([, tool]) => {
|
|
const fresh = current.get(tool.name);
|
|
return fresh && permits(tool) && permits(tool, currentWorkMode) &&
|
|
permits(fresh) && permits(fresh, currentWorkMode);
|
|
});
|
|
if (hasSchemaTool) {
|
|
const selected = authorized.find(([name]) => name === args.schemaTool);
|
|
if (!selected) throw new Error("assigned_mcp_tool_unknown: schemaTool must match an authorized inputSchemaRef. Omit schemaTool to search assigned tools, then use an exact inputSchemaRef from the results.");
|
|
const schema = JSON.stringify(selected[1].parametersSchema);
|
|
const end = schemaOffset + schemaChunkCharacters;
|
|
return {
|
|
name: selected[0], schemaJson: schema.slice(schemaOffset, end),
|
|
nextSchemaOffset: end < schema.length ? end : null,
|
|
};
|
|
}
|
|
// Keep offsets anchored to the pinned catalog while fresh grants only
|
|
// decide which entries are returned from that stable sequence.
|
|
const authorizedNames = new Set(authorized.map(([name]) => name));
|
|
const matches = [...tools].filter(([, tool]) => permits(tool)).filter(([name, tool]) =>
|
|
terms.every(term => `${name} ${tool.displayName} ${tool.description}`.toLowerCase().includes(term)))
|
|
.sort(([a], [b]) => a < b ? -1 : a > b ? 1 : 0);
|
|
const page: ToolDefinition[] = [];
|
|
let nextOffset = offset;
|
|
while (nextOffset < matches.length && page.length < limit) {
|
|
const [name, tool] = matches[nextOffset]!;
|
|
if (!authorizedNames.has(name)) {
|
|
nextOffset += 1;
|
|
continue;
|
|
}
|
|
let next = definition(name, tool);
|
|
if (Buffer.byteLength(JSON.stringify([next]), "utf8") > searchPageBytes) {
|
|
// One large schema must not block browsing the tools after it. Expose
|
|
// a reference whose schema can be fetched in bounded, reauthorized
|
|
// chunks, without weakening the gateway's argument validation.
|
|
next = { name, description: String(next.description).slice(0, 2000), inputSchemaRef: name };
|
|
}
|
|
if (Buffer.byteLength(JSON.stringify([...page, next]), "utf8") > searchPageBytes) {
|
|
break;
|
|
}
|
|
page.push(next);
|
|
nextOffset += 1;
|
|
}
|
|
return { tools: page, nextOffset: nextOffset < matches.length ? nextOffset : null };
|
|
}
|
|
|
|
return {
|
|
definitions(fits?: (definitions: ToolDefinition[]) => boolean): ToolDefinition[] {
|
|
const direct = [...tools].filter(([, tool]) => permits(tool)).map(([name, tool]) => definition(name, tool));
|
|
if (input.delivery !== "on_demand" && (!fits || fits(direct))) return direct;
|
|
const compact = structuredClone(ON_DEMAND_TOOLS);
|
|
if (fits && !fits(compact)) throw new Error("assigned_mcp_tool_catalog_capacity_exceeded");
|
|
return compact;
|
|
},
|
|
has(name: string): boolean {
|
|
return tools.has(name) || name === SEARCH_TOOL || name === CALL_TOOL;
|
|
},
|
|
async execute(call: { tool: string; arguments: unknown }, currentWorkMode?: WorkMode): Promise<unknown> {
|
|
if (call.tool === SEARCH_TOOL) return search(call.arguments, currentWorkMode);
|
|
let name = call.tool;
|
|
let parameters = call.arguments;
|
|
if (name === CALL_TOOL) {
|
|
const args = object(parameters);
|
|
if (typeof args.name !== "string") throw new Error("assigned_mcp_tool_invalid_arguments");
|
|
name = args.name;
|
|
parameters = object(args.arguments);
|
|
}
|
|
const descriptor = tools.get(name);
|
|
if (!descriptor) throw new Error("assigned_mcp_tool_unknown");
|
|
if (!permits(descriptor) || !permits(descriptor, currentWorkMode)) throw new Error("paperclip_runner_tool_mode_denied");
|
|
// Reauthorize through the existing gateway on every call. Discovery is
|
|
// not a grant: revocation, policy, approval, and audit remain server-owned.
|
|
const result = await input.gateway.executeTool({
|
|
gatewayPublicId: input.gatewayPublicId,
|
|
sessionToken: input.bearerToken,
|
|
tool: descriptor.name,
|
|
parameters,
|
|
});
|
|
if (result.status !== "completed" && result.status !== "replayed") {
|
|
throw new Error("assigned_mcp_tool_execution_incomplete");
|
|
}
|
|
// The gateway normally throws provider errors. Keep a malformed or older
|
|
// provider response from being reported as successful through PRP.
|
|
if (result.result && typeof result.result === "object" && "isError" in result.result && result.result.isError === true) {
|
|
throw new ToolGatewayHttpError(502, "The assigned tool returned an error.", "tool_error");
|
|
}
|
|
return result.result;
|
|
},
|
|
};
|
|
}
|