Composio powers 1000+ toolkits, tool search, context management, authentication, and a sandboxed workbench to help you build AI agents that turn intent into action.

next
md 47 lines 2.38 KB
Rendered Raw
1 # Docs Knowledge Base — removed / parked content
2
3 A dumping ground for content removed from the published docs during the docs
4 overhaul. Nothing here is rendered. Keep it so we don't lose nuance we may want
5 to reintroduce (e.g. in deeper guides, FAQs, or the LLM-facing `.md` output).
6
7 Each entry: **what it was**, **where it came from**, **why removed**, **the content**.
8
9 ---
10
11 ## 1. "Native Tools vs MCP" decision framing
12
13 **Where:** `content/docs/native-tools-vs-mcp.mdx` (whole page) + the native/MCP
14 tab split in `quickstart.mdx`.
15
16 **Why removed:** We decided to stop presenting native-vs-MCP as a decision the
17 user has to make up front. Default everyone to **native tools**; MCP becomes an
18 opt-in documented on a single "Using sessions via MCP" page. This removes a fork
19 in the road from the first-run experience.
20
21 **Parked content — the comparison table + token-cost argument:**
22
23 > **Native tools** give your LLM tool schemas as function definitions. Composio
24 > formats them for your specific framework (OpenAI, Anthropic, Vercel AI, etc.)
25 > through provider packages.
26 >
27 > **MCP** exposes tools through the Model Context Protocol. Any MCP-compatible
28 > client can connect to a Composio MCP server URL. No provider packages needed.
29 >
30 > | | Native tools | MCP |
31 > |---|---|---|
32 > | **Setup** | Provider package for your framework | SDK or just a URL |
33 > | **Intercepting tool calls** | Yes, you can log, retry, or require approval before each call | Limited, depends on what the MCP client supports |
34 > | **Context window** | You control what's loaded | Client loads all tools the server exposes |
35 > | **Latency** | SDK calls Composio API directly | MCP protocol adds overhead for tool list discovery and each execution |
36 >
37 > With native tools, you choose exactly which schemas enter your LLM's context.
38 > With MCP, the client pulls the full tool list from the server. A
39 > [5-server setup can consume ~55K tokens](https://www.anthropic.com/engineering/advanced-tool-use)
40 > before the conversation starts. If you're working with many tools, native
41 > tools give you more control over that cost.
42
43 **Note:** the token-cost framing is slightly inaccurate for Composio sessions
44 since session.tools() returns *meta tools* (search/execute) rather than the full
45 tool list — so the "55K tokens" client-loads-everything point applies to naive
46 MCP, not necessarily to a Composio session over MCP. Worth a cleaner write-up if
47 reintroduced.